Technical Summary: Dormitory Peer Effects on Physical Fitness Decline in Chinese Undergraduates

Problem Statement

Physical fitness deterioration among college students is a documented global public health issue, with Chinese undergraduates experiencing a significant decline in cardiorespiratory endurance and overall fitness pass rates over the past decade. While individual determinants such as exercise knowledge and facility availability are well-studied, the role of the social environment—specifically peer influences within residential dormitory settings—remains under-explored in the Chinese context. Existing literature often relies on single-cohort cross-sectional designs that cannot disentangle age, period, and cohort effects, or focuses on Western contexts where roommate selection is self-driven and assignments change frequently. The Chinese university system, characterized by administrative assignment of 6–8 students per room for four years without self-selection, presents a unique natural experiment to investigate how non-random peer networks shape health trajectories.

Methodology

This study employed a dual-cohort longitudinal design tracking 4,157 students (2021 cohort: n=1,836; 2022 cohort: n=2,321) across four academic years at a comprehensive university in Shandong Province, China.

Data Collection and Measures

• Physical Fitness: Assessed annually using the National Student Physical Fitness Standard (NSPFS), a weighted composite (0–120 scale) of seven components: BMI, vital capacity, 50m dash, sit-and-reach, standing long jump, sit-ups/pull-ups, and endurance running.
• Dormitory Assignment: Students were assigned to rooms by an administrative algorithm based solely on college and major, with less than 3% roommate turnover over four years, minimizing homophily-based selection bias.
• Covariates: Sex, age, cohort year, college, major category, and urban/rural origin.

Analytic Strategy

The study utilized a "mixed-methods analytics" approach integrating five distinct stages:

1. Descriptive Profiling: Analysis of means, standard deviations, and effect sizes.
2. Hierarchical Linear Modeling (HLM): Three-level models (occasion within student within dormitory) with random intercepts and slopes to quantify dormitory clustering (Intraclass Correlation Coefficients, ICCs).
3. Latent Growth Curve Modeling (LGCM): Unconditional linear and quadratic models to determine individual fitness trajectories.
4. K-means Cluster Analysis: Applied to standardized dormitory-level fitness vectors to identify natural dormitory typologies without a priori categories.
5. Machine Learning (Gradient Boosting): XGBoost models with SHAP (SHapley Additive exPlanations) values to rank the relative importance of predictors (dormitory membership vs. individual factors) for senior-year fitness outcomes.
6. Age-Period-Cohort (APC) Decomposition: Used the Yang-Land intrinsic estimator to separate maturation, temporal (e.g., pandemic), and cohort effects.

Key Results

1. Substantial Dormitory Clustering

Significant clustering of fitness outcomes was observed within dormitories, with ICCs ranging from 0.152 to 0.227 across grades and cohorts. The variance decomposition indicated that 18.6% of total fitness variance was attributable to between-dormitory differences, peaking during the sophomore year.

2. Four Dormitory Typologies

K-means clustering identified four distinct dormitory archetypes:

• Synchronized Decline (38%): Characterized by high peer synchronization (84.2%) and the largest fitness decline (-12.4 points).
• Divergent (27%): Exhibited the lowest synchronization (61.7%) and heterogeneous change patterns.
• Improvement (19%): Showed net fitness gains (+5.6 points).
• Resilient (16%): Maintained baseline fitness with minimal decline (-0.8 points).

3. Peer Synchronization Dynamics

The overall peer synchronization rate (concordant fitness change direction among roommates) was 73.4%, significantly exceeding the 50% chance expectation. Synchronization was highest during the freshman-to-sophomore transition (76.8%) and was amplified when roommates had similar baseline fitness levels.

4. Predictive Importance of Social Environment

SHAP-based feature importance rankings revealed that dormitory cluster membership (mean |SHAP| = 4.82) was the second-strongest predictor of senior-year fitness outcomes, surpassing individual biological factors such as BMI (3.47), sex (2.95), and college major (1.88). Only baseline fitness (8.31) was a stronger predictor.

5. Age-Period-Cohort Decomposition

APC analysis attributed 62% of systematic variance to age-related maturation, 23% to period effects (primarily COVID-19 lockdowns), and 15% to cohort differences. The 2022 cohort showed a steep initial decline followed by partial recovery, suggesting resilience when environmental constraints are removed.

6. Demographic Disparities

Rural students entered with lower baseline fitness and were overrepresented in "synchronized-decline" dormitories (42.3% vs. 35.1% for urban students), despite experiencing comparable peer influence mechanisms. Male students exhibited steeper decline rates, widening the gender fitness gap from 6.2 to 9.8 points over four years.

Key Contributions

The paper claims four primary contributions to the literature:

1. Methodological: The integration of a dual-cohort longitudinal design with multi-method analytics (HLM, LGCM, K-means, and SHAP-based machine learning) allows for the formal separation of age, period, and cohort effects and the ranking of social versus individual predictors.
2. Analytical: The identification of a four-cluster dormitory taxonomy provides a data-driven framework for risk stratification, moving beyond binary "peer effect" existence to characterize the nature of peer influence (synchronized decline vs. improvement).
3. Theoretical: The study empirically adjudicates between Social Cognitive Theory, Social Contagion Theory, and Ecological Systems Theory, finding evidence for all three: directional influence (improvement clusters), homogenizing convergence (synchronized decline), and contextual moderation (divergent clusters).
4. Practical: The findings offer actionable intelligence for dormitory-based interventions, suggesting that targeting the "freshman-sophomore transition window" and leveraging existing dormitory structures could be a scalable strategy for population-level fitness promotion.

Significance and Claims

The authors assert that dormitory peer effects are a substantial and policy-relevant determinant of physical fitness trajectories in Chinese higher education. The study challenges traditional intervention hierarchies that prioritize individual biological factors (like BMI) by demonstrating that the social-environmental context (dormitory cluster membership) is a more powerful predictor of long-term fitness outcomes.

The paper emphasizes that the Chinese dormitory system, with its administrative assignment and four-year stability, creates conditions for peer influence that may exceed those in Western contexts. Consequently, the observed effect sizes (ICCs of ~0.20) likely represent an upper bound for dormitory peer effects. The authors conclude that interventions should shift from targeting individual students to modifying the social composition or normative climate of dormitories, particularly by identifying and supporting "improvement" and "resilient" clusters to harness positive contagion effects.

The study acknowledges limitations, including its observational nature (precluding definitive causal claims despite the natural experiment design), the reliance on a composite fitness score which may mask component-specific effects, and the lack of intermediate behavioral data (e.g., exercise frequency). However, the convergence of evidence across multiple analytical methods strengthens the validity of the central conclusion: dormitory peer effects are measurable, substantial, and actionable.