Study design and participants
We conducted a cross-sectional survey from April 24th to 30th, 2024, at five universities in China. The authors initially developed a preliminary questionnaire based on a literature review. This questionnaire was then refined according to feedback from a pilot study. The final version was uploaded to “Sojump” (www.sojump.com), a professional online platform for designing, distributing, collecting, and analyzing questionnaires [23]. To distribute the survey, we enlisted 85 college students who shared a quick response (QR) code linked to our survey in their online class chat groups. We described the study as investigating the mental health and lifestyle habits of college students, without disclosing the specific research questions during participant recruitment. Participation was voluntary, and no rewards were given for completing the questionnaire. Informed consent forms were signed by participants before initiating the survey. This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Review Committee of Southwest University (SWUETH20230912003).
Inclusion criteria: (i) university students aged ≥ 18 years; (ii) had used short videos in the last month. Exclusion criteria: (i) lack of informed consent; (ii) unfinished questionnaires; (iii) individuals who failed verification tests (to confirm that the questionnaire was completed carefully by individuals rather than by a machine, program, or random filling).
Instruments and measurements
Depression
The Patient Health Questionnaire (PHQ-9) was utilized to assess the levels of depression [24]. This scale comprises nine questions, and the PHQ-9 aligns with the criteria for a major depressive episode as outlined in the DSM-IV. Respondents rate the frequency of their depressive symptoms over the past two weeks, with scores ranging from 0 (not at all) to 3 (nearly every day). We utilized a Chinese version of the PHQ-9, which demonstrated strong internal consistency in our study (Cronbach’s α = 0.843).
Short video addiction
In this study, we employed the Short-Form Video App Addiction Scale to assess addiction to short videos [1]. This scale comprises six self-report items designed to assess individuals’ addictive behaviors linked to their usage of short videos. Participants were asked to consider their recent experiences and feelings regarding short video usage, rating on a 7-point Likert scale (1 = strongly disagree and 7 = strongly agree). We utilized a Chinese version of the scale, which demonstrated strong internal consistency in our study (Cronbach’s α = 0.857).
Short video usage during daytime, nighttime, study time, leisure time, and overall usage
We referred to previous publications and assessed various short video usage behaviors through frequency and duration metrics [6, 25]. For instance, the overall short video usage was evaluated using the following questions:
Frequency: “In the last month, how often did you use short video each week?” Responses were collected across 8 categories: 1 = never, 2 = an average of 1 day per week, 3 = an average of 2 days per week, 4 = an average of 3 days per week, 5 = an average of 4 days per week, 6 = an average of 5 days per week, 7 = an average of 6 days per week, and 8 = an average of 7 days per week.
Duration: “On days when you used short video, how long did you typically use them each day?” Responses were gathered across 8 categories: 1 = less than 30 min, 2 = 31 to 60 min, 3 = 61 to 90 min, 4 = 91 to 120 min, 5 = 121 to 150 min, 6 = 151 to 180 min, 7 = 181 to 240 min, and 8 = more than 240 min.
The product of the frequency and duration scores was calculated as the outcome. Other short video usage behaviors were also assessed using similar methods, with the only difference being the specific questions asked. The detailed information is presented in Supplement 1.
Physical activity
We drew on the design used in the UK Biobank for measuring physical activity (e.g., In a typical week, on how many days did you engage in 10 min or more of moderate physical activities) and insights from related publications [6, 26], assessing physical activity in this study through frequency and duration. The specific question is as follows:
Frequency: “In the last month, how often did you engage in moderate or higher intensity physical activity each week?” Responses were gathered across 8 categories: 1 = never, 2 = an average of 1 day per week, 3 = an average of 2 days per week, 4 = an average of 3 days per week, 5 = an average of 4 days per week, 6 = an average of 5 days per week, 7 = an average of 6 days per week, and 8 = an average of 7 days per week.
Duration: “On days when you participated in moderate or higher intensity physical activity, how long did you typically spend each day?” Responses were collected across 8 categories: 1 = less than 30 min, 2 = 31 to 60 min, 3 = 61 to 90 min, 4 = 91 to 120 min, 5 = 121 to 150 min, 6 = 151 to 180 min, 7 = 181 to 240 min, and 8 = more than 240 min.
The product of the frequency and duration scores was calculated as the outcome. The complete questions are provided in Supplement 1.
Statistical analysis
Demographic information and variables were presented as means with standard deviations (SD) or numbers with percentages. Due to the non-normal distribution of the data, Spearman’s rank-order correlation was employed to explore general correlations between variables. Structural equation modeling (SEM) was utilized to investigate the hypothesized directional paths within the conceptual frameworks. Following the guidance of RP Bagozzi and Y Yi [27], our sample size of 1172 exceeded the recommended size of twice the number of model parameters. Variance Inflation Factor (VIF) values less than 5.0 were used to indicate the absence of multicollinearity [28]. Based on this threshold, none of the independent variables exhibited multicollinearity (VIF < 3.0). Due to the non-normality of the data, we employed an asymptotically distribution-free (ADF)/weighted least squares (WLS) estimator for SEM [29, 30], as this estimator is suited for non-normal data when the sample size is large (sample size: 1000 to 5000 and at least 10 times the number of estimated parameters) [31]. The bootstrap method with 10,000 replications was utilized to compute corresponding standard errors and confidence intervals for all paths [32,33,34]. Based on the ADF/WLS estimator, we evaluated the goodness of fit using the following indices [30, 35]: standardized root mean square residual (SRMR) < 0.08; Tucker–Lewis index (TLI) > 0.95; goodness-of-fit index (CFI) > 0.95, and root mean square error of approximation (RMSEA) < 0.05. We opted not to employ the χ2 test due to its susceptibility to sample size effects and violations of the multivariate normality assumption [36,37,38]. An indirect effect (i.e., a product of coefficients for the constituent links) that significantly exceeded zero was evidence of mediation [39, 40].
All statistical analyses were performed using SPSS 26.0 and AMOS 23.0 software (SPSS Inc., Chicago, IL, USA), with p-values < 0.05 considered statistically significant [41, 42].
link