Answers are grounded in this paper and cite their sources.
Academic stress has been consistently associated with depressive symptoms among adolescents. Understanding whether physical exercise can buffer its negative effects is important for promoting adolescent mental health. Using data from the 2020 China Family Panel Studies, this study examined an unweighted analytical sample of 2789 school-attending adolescents aged 16–18 years. Hierarchical linear regression and a series of psychometric, diagnostic, and sensitivity analyses were employed to investigate the association between academic stress and depressive symptoms, the moderating role of exercise frequency, and gender differences in this association. Extreme Gradient Boosting (XGBoost) and SHapley Additive exPlanations (SHAP) analyses were also conducted as exploratory complementary analyses. After adjusting for demographic covariates, academic stress was significantly and positively associated with depressive symptoms (B = 0.542, 95% CI [0.444, 0.640], β = 0.200, p < 0.001), whereas exercise frequency was significantly and negatively associated with depressive symptoms (B = −0.138, 95% CI [−0.195, −0.081], β = −0.089, p < 0.001), with the main-effects model explaining 5.9% of the variance (R2 = 0.059). The interaction between academic stress and exercise frequency was significantly negative (B = −0.060, 95% CI [−0.113, −0.007], β = −0.041, p = 0.026, full model R2 = 0.061), indicating that higher exercise frequency was associated with a weaker positive relationship between academic stress and depressive symptoms. However, the interaction explained only an additional 0.2% of the variance (Δ R2 = 0.002), suggesting a modest buffering association that remained consistent across weighted, cluster-robust, Tobit, negative binomial, and multiple-imputation sensitivity models. An exploratory, non-preregistered three-way interaction model (R2 = 0.063, Δ R2 = 0.002, B = −0.080, 95% CI [−0.158, −0.002], p = 0.046) suggested gender differences in this moderating pattern, with a statistically significant conditional buffering association observed among male adolescents (B = −0.102, 95% CI [−0.178, −0.026], p = 0.009) but not among female adolescents (B = −0.022, 95% CI [−0.095, 0.051], p = 0.552). In the exploratory machine learning analyses, continuous XGBoost regression explained no additional out-of-sample variance beyond linear regression (R2 = 0.061 vs. 0.060), and TreeSHAP interaction values confirmed a small interaction magnitude. Overall, these findings are consistent with the stress-buffering framework in the context of adolescent academic stress and describe associations in this cross-sectional Chinese sample; however, the interaction added little explanatory variance, and the cross-sectional observational design precludes causal inference.
No takes yet. Share an insight, caveat, or question.
Du et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: