24-h movement behaviors and internet addiction symptoms among Chinese university students: a cross-sectional compositional isotemporal substitution analysis using generalized additive modeling
Background Physical activity, sedentary behavior, and sleep compete within a finite 24-h day, yet their associations with internet addiction symptoms are often examined separately. Objective To examine associations between the four-part 24-h movement composition and internet addiction symptoms among Chinese university students and to estimate model-predicted differences for hypothetical sedentary behavior (SB)-to-moderate-to-vigorous physical activity (MVPA) reallocations using compositional isotemporal substitution and generalized additive modeling. Methods In this cross-sectional study, 986 students from 42 classes at one university in Zhengzhou completed the adapted 26-item Chinese Internet Addiction Scale (CIAS) and a 7-day 24-h movement-behavior questionnaire. Self-reported time in bed with the intention to sleep, SB, light physical activity, and MVPA were closed to 1,440 min/day and transformed into isometric log-ratio coordinates. Adjusted compositional regression with class-clustered robust standard errors, a generalized additive model (GAM) with a penalized MVPA smooth, and compositional isotemporal substitution were used; uncertainty was assessed with 5,000 stratified class-level bootstrap resamples. Results The overall composition was associated with CIAS total score [ F (3, 40.5) = 12.45, p < 0.001]. A higher relative MVPA share was associated with lower CIAS total score [β = −7.21, 95% CI [−10.15, −4.27]). The GAM fit modestly better than the linear model (ΔAIC = −4.2; explained deviance, 15.6 vs. 14.2%; nested F -test for departure from linearity, F = 3.42, p = 0.028). Relative to the sample compositional center, hypothetical 15-, 30-, and 45-min/day SB-to-MVPA reallocations yielded closely aligned predicted CIAS differences in the GAM (−0.85 [95% CI −1.40, −0.56], −1.50 [−2.24, −0.97], and −2.02 [−3.10, −1.11] points) and the linear compositional model (−0.88 [−1.24, −0.52], −1.56 [−2.19, −0.92], and −2.12 [−2.98, −1.26] points). While both models produced curved trajectories on the minute scale due to the log-ratio transformation, the GAM captured modest additional tail attenuation and wider uncertainty above 30 min/day; no precise threshold or plateau was identified. Findings remained consistent across sensitivity analyses. Conclusions In this cross-sectional sample, greater relative MVPA and model-based SB-to-MVPA reallocations were associated with lower internet addiction symptom scores. Causal effects and optimal reallocation amounts remain undetermined.
Authors
- Jixin Wang (ORCID: https://orcid.org/0000-0002-8847-7481)
- Zhihao Du
- Danning Wang
- Shuai Liu
- Guan Yang
- Qi Zhou
Institutions
- Shinhan University (KR)
- Qingdao Municipal Hospital (CN)
- Changzhou University (CN)
- South China University of Technology (CN)
Publication Details
- Journal
- Frontiers in Public Health
- Published
- 2026-11-20
- DOI
- https://doi.org/10.3389/fpubh.2026.2017503
- Primary Topic
- Physical Activity and Health
- Type
- article
- Field-Weighted Citation Impact
- 0.00