Longitudinal associations among social support, physical activity, and problematic eating behavior in adolescents: a three-wave random-intercept cross-lagged panel study

Background Unhealthy dietary habits and inactive lifestyles contribute to weight gain during adolescence. Nevertheless, how social support, physical activity, and problematic eating behavior relate over time remains poorly understood, especially in separating enduring differences between individuals from fluctuations within individuals. We therefore conducted a three-wave longitudinal study with adolescents and used random-intercept cross-lagged panel modeling to examine these relationships. Methods We recruited a convenience sample of 671 students from 10 secondary schools in Jiangxi, Zhejiang, Fujian, and Guangdong provinces. Participants completed three waves of assessments at 6-month intervals. We measured social support, physical activity, and problematic eating behavior with the Social Support Rating Scale, the Physical Activity Rating Scale, and the Short Form of the Sakata Eating Behavior Scale, respectively. We analyzed data in SPSS 26.0 and Mplus 8.3. The RI-CLPM served as the primary analytic framework for separating within-individual fluctuations from enduring between-individual differences. We fitted a conventional CLPM as a complementary model to examine cross-wave associations. Results In the primary RI-CLPM, most within-person cross-lagged associations among the three constructs were not statistically significant, although one path was nominally significant ( β = −0.113, p = 0.049). The random intercepts were strongly correlated (|r| = 0.871–0.887). In the supplementary conventional CLPM, social support was positively associated with subsequent physical activity ( β = 0.246–0.254, p < 0.001), whereas social support and physical activity were inversely associated with subsequent problematic eating behavior ( β = −0.260 to −0.126, p < 0.01). We also observed a significant indirect association between T1 social support and T3 problematic eating behavior through T2 physical activity (standardized indirect effect = −0.032, p < 0.01). However, these conventional CLPM findings were not replicated at the within-person level in the RI-CLPM and may partly reflect stable between-person differences rather than within-person temporal dynamics. Conclusion Under the RI-CLPM, most cross-lagged links did not reach significance at the within-person level, with only one path showing nominal significance. By contrast, the conventional CLPM produced significant cross-lagged associations along with an indirect association. Yet the RI-CLPM did not replicate these patterns at the within-person level. Such a mismatch underscores why research on adolescent health behavior must separate stable between-person associations from within-person processes.