Temporal Dynamics of Influenza-Associated Anxiety Symptom Linguistic Markers on Weibo (2023-2024): Observational Study.

Driven by visual culture and health awareness, exercise has become a core lifestyle (). In China, physical activity participation and the fitness market are growing, yet the gap between exercise intention and actual behavior remains wide: most adults desire to exercise, but long-term adherence is low (). Early dropout is common, highlighting the challenge of sustaining behavior. This raises key questions: Are exercisers a homogeneous group? Do adherence mechanisms differ? Existing research spans health psychology (self-efficacy, goal setting; ()), sociology (group norms, social capital; ()), and consumer culture (body anxiety, appearance management; ()). However, three limitations persist. First, variable-centered approaches overlook individual heterogeneity. Structural equation modeling and regression assume uniform causal mechanisms (, ), but social support may buffer stress for some and burden others. Second, theoretical integration is lacking. Appearance anxiety (), social support (), and self-determined motivation () are often studied separately; a unified framework is missing. Third, precision interventions lack grounding. Homogeneity-based strategies ignore individual differences. Moreover, appearance anxiety, pervasive in digital and visual cultures, is often seen as a negative stressor, yet many highly dissatisfied individuals maintain long-term exercise—a paradox requiring examination of boundary conditions. To address these issues, we shift from a variable-centered to a person-centered perspective (). Latent Profile Analysis (LPA) identifies distinct latent categories based on multidimensional response patterns (). While person-centered methods have been used in exercise psychology (e.g., (, )), they often focus on single dimensions, lacking integration of appearance anxiety, social support, and self-driven factors. Compared with variable-centered methods, latent profile analysis is more suitable for testing the heterogeneity hypothesis of this study. Regression or traditional SEM estimates population-averaged effects, implicitly assuming that different individuals follow the same parametric structure. However, under the PSM framework, pressure, support, and motivational resources may co-occur in different configurations, thereby forming multiple psychological profiles characterized by “the same variables, but different configurations.” In such cases, population-averaged effects may be “canceled out” (for instance, appearance anxiety may be positively associated with persistence in one subgroup, while being uncorrelated or negatively associated in another), thus obscuring key differences. LPA identifies latent subgroups at the level of joint distribution of multiple indicators, allowing this study to not only answer “which variables are associated with persistence,” but also to determine “which types of individuals are more likely to exhibit certain resource combinations in relation to persistence.”