Relative associations of physical activity, social support, and self-control with resilience among vocational college...

Vocational college students face substantial academic, occupational, and social transition demands, yet the relative importance of modifiable behavioral and psychosocial resources associated with resilience remains unclear.

This multisite cross-sectional study examined physical activity, social support, self-control, and resilience among 2,251 students recruited by convenience sampling from eight vocational colleges in Jiangsu and Henan Provinces, China. All constructs were assessed using self-report measures. Spearman correlations and covariate-adjusted linear regression were used to compare bivariate and independent associations. Exploratory parallel indirect-association models statistically decomposed the cross-sectional physical activity-resilience association through social support and self-control. Complementary out-of-sample predictive benchmarking compared Ridge regression with gradient boosting, random forest, and support vector regression using repeated nested cross-validation. A personal-resilience outcome was examined as a construct-overlap sensitivity analysis.

Social support showed the strongest adjusted association with resilience (standardized β = 0.587, 95% CI [0.553, 0.621]), followed by self-control (β = 0.344, 95% CI [0.312, 0.376]) and physical activity (β = 0.047, 95% CI [0.018, 0.076]). Positive indirect associations were observed through social support (0.0725, 95% CI [0.0431, 0.1009]) and self-control (0.0653, 95% CI [0.0486, 0.0831]). Ridge regression achieved the numerically highest mean held-out R ² (0.679), but none of the nonlinear models showed statistically supported improvement beyond the linear benchmark; ordinary least-squares sensitivity analysis produced nearly identical performance (R ² = 0.678). The broad pattern remained similar when personal resilience was used.

In this sample, social support and self-control were substantially stronger independent correlates of resilience than physical activity, while additional nonlinear model complexity offered little out-of-sample predictive gain.