Between-class variation and compositional correlates of physical fitness: a multilevel analysis of annual surveillance records from one Chinese university.

This retrospective study quantified between-class variation in university students' physical fitness and examined individual- and class-level correlates, using annual surveillance records collected from 2020 to 2024 at one public university in Sichuan Province, China. Records were nested within class-year units (a specific administrative class observed in a specific survey year). Although no psychosocial constructs were measured directly, quantifying between-class variation is a necessary first step for health-psychology research on how peer norms, motivational climate, and collective efficacy may relate to student fitness; the study is therefore hypothesis-generating rather than a test of mechanisms. A total of 106,461 de-identified records from 2,591 class-year units were analyzed (44,240 male, 62,221 female; mean age 20.67 ± 1.54 years; mean BMI 21.64 ± 3.52 kg/m2). Multilevel linear models were fitted, with calendar year (2020-2024) entered as a categorical fixed-effect set in the primary specification. Between-class variation was substantial (null-model intraclass correlation coefficient [ICC] = 0.178; adjusted ICC after individual and calendar-year adjustment = 0.151). Higher within-class BMI, older age, and male sex were associated with lower fitness (within-class BMI β = -0.528, p < 0.001; sex [male vs. female] β = -4.174, p < 0.001); second- and fourth-year students scored higher than first-year students (β = 0.803 and 1.913). At the class level, a higher proportion of male students (β = 1.090, p < 0.001) was positively associated with fitness, whereas larger class size (β = -0.012, p = 0.021; a practically negligible 0.24-point difference across a 20-student gap) and higher mean class BMI (β = -0.906, p < 0.001, the largest class-level association) were associated with lower fitness. Standardized coefficients confirmed that sex and within-class BMI were the strongest individual-level correlates; the fixed effects explained 15.4% of the variance (marginal R 2 = 0.154; conditional R 2 = 0.282). A random-slope extension showed that the within-class BMI-fitness association varied across classes (likelihood-ratio χ2 = 463.07, df = 2, p < 0.001), but this heterogeneity was modest in practical terms (slope SD ≈ 0.27; ~95% of class-specific slopes between -1.05 and +0.01). Findings provide evidence of meaningful class-level variation-not a test of mechanisms-and indicate that analyses and interventions targeting university physical fitness should account for the clustered structure of the data.