Adolescent Eating and Exercise Links Fade When Tracking Individual Changes

A three-wave longitudinal study tracking secondary school students across four Chinese provinces has found that commonly observed longitudinal links connecting social support, physical activity, and problematic eating behavior largely fail to hold when within-person variations are separated from stable differences between individuals. Published in Frontiers in Nutrition by a research team including Yuan Yuqing, Chang Hu, Wen Zhang, and Yafei Chen, the investigation evaluated how interpersonal support, exercise, and disordered dietary patterns interrelate over time during adolescence. By comparing a primary random-intercept cross-lagged panel model against a traditional cross-lagged modeling approach, the researchers demonstrated that standard analytic techniques can identify statistically significant longitudinal associations that actually reflect enduring baseline traits rather than dynamic changes unfolding within individual adolescents.

The research was designed to address an ongoing challenge in adolescent behavioral health, where unhealthy dietary habits and physically inactive lifestyles contribute substantially to weight gain and related complications. While prior literature has frequently linked social support and physical exercise to healthier eating patterns, the precise temporal ordering and nature of these constructs over time have remained poorly understood. Specifically, researchers have struggled to disentangle whether adolescents who report greater social support and activity engage in less problematic eating because of enduring, stable differences that distinguish one person from another, or whether positive shifts within a given individual over time actively drive subsequent behavioral improvements.

To address this question empirically, the researchers recruited a convenience sample of 671 students drawn from 10 secondary schools located across Jiangxi, Zhejiang, Fujian, and Guangdong provinces. The participating adolescents completed structured assessments across three distinct measurement waves, with each wave spaced at six-month intervals to monitor behavioral trajectories across an extended developmental window. The multi-province school sample provided a substantial cohort for evaluating longitudinal interactions across repeated developmental observations.

The investigation utilized three established self-report psychometric scales to measure the targeted behavioral dimensions across each evaluation cycle. Social support was quantified using the Social Support Rating Scale, while physical activity levels were documented through the Physical Activity Rating Scale. Problematic eating behavior was evaluated using the Short Form of the Sakata Eating Behavior Scale. To evaluate the resulting longitudinal dataset, the research team conducted computational and statistical procedures utilizing SPSS version 26.0 alongside structural equation modeling routines in Mplus version 8.3.

The analytical strategy centered on comparing two structural panel frameworks to understand how modeling assumptions shape empirical conclusions. The random-intercept cross-lagged panel model served as the primary analytic framework, specifically implemented because it explicitly partitions enduring, stable between-individual differences from prospective, within-individual fluctuations across time. For comparative evaluation, the team also fitted a supplementary conventional cross-lagged panel model, which examines cross-wave associations across time points without separating stable inter-individual differences from intra-individual changes.

In the primary random-intercept cross-lagged panel model, the researchers observed that most within-person cross-lagged associations among social support, physical activity, and problematic eating behavior were not statistically significant over the six-month intervals. Across the within-person paths evaluating whether changes in one domain predicted subsequent changes in another, only a single cross-lagged pathway achieved nominal statistical significance, yielding a standardized coefficient of beta = -0.113 (p = 0.049). In contrast, the model’s random intercepts exhibited strong correlations ranging from |r| = 0.871 to 0.887, demonstrating that the vast majority of shared variance across social support, exercise, and problematic eating was concentrated in stable, trait-like differences between individual students.

When the researchers evaluated the same longitudinal dataset using the supplementary conventional cross-lagged panel model, the results displayed a strikingly different pattern of statistical significance. In the conventional model, higher social support was positively associated with subsequent physical activity across waves, with standardized regression coefficients ranging from beta = 0.246 to 0.254 (p < 0.001). Concurrently, both social support and physical activity exhibited inverse associations with subsequent problematic eating behavior, producing coefficients ranging from beta = -0.260 to -0.126 (p < 0.01). The conventional model also identified a statistically significant indirect pathway connecting Time 1 social support to Time 3 problematic eating behavior mediated through Time 2 physical activity, with a standardized indirect effect of -0.032 (p < 0.01).

Crucially, these apparent prospective effects and indirect pathways identified by the conventional cross-lagged panel model were not replicated at the within-person level in the primary random-intercept model. The authors concluded that the significant cross-wave links and mediation effects detected by standard modeling methods may partly reflect stable, preexisting differences between individuals rather than true temporal dynamics operating within individuals over time.

Subject of Research: Longitudinal relationships among social support, physical activity, and problematic eating behavior in adolescents

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

Article References: Yuan, Y., Hu, C., Zhang, W., Chen, X., Chen, Y., & Zhang, C. (2026). Longitudinal associations among social support, physical activity, and problematic eating behavior in adolescents: a three-wave random-intercept cross-lagged panel study. Frontiers in Nutrition, 13, 1948759. https://doi.org/10.3389/fnut.2026.1948759

Image Credits: AI Generated

DOI: 10.3389/fnut.2026.1948759

Keywords: adolescent health, social support, physical activity, problematic eating behavior, random-intercept cross-lagged panel model, cross-lagged panel model, longitudinal study, within-person dynamics, between-person differences, Social Support Rating Scale, Physical Activity Rating Scale, Short Form Sakata Eating Behavior Scale, secondary school students, SPSS analysis, Mplus modeling, dietary habits, temporal associations, mediational pathways