Integrating physical, emotional, and social changes to examine their associations with self-efficacy among college students engaging in physical activity

Introduction

College students face increasing physical and psychological challenges, and the multidimensional association between physical activity and well-being warrants clarification. This study tested an integrative model linking physical activity to perceived physical, emotional, and social relationship changes and self-efficacy, and examined variation by sex and training time.

Methods

A cross-sectional survey was conducted among 1,082 full-time university students in China (inclusion: basic engagement in physical activity; exclusion: major physical or mental illness). Validated questionnaires were used to assess physical activity, self-efficacy, health status, and well-being, together with study-developed subscales for perceived physical, emotional, and social relationship changes. Internal consistency and construct validity were evaluated in the study sample, and the intended measurement structure was supported. Correlation analyses and structural equation modelling (SPSS/AMOS) examined model-based associations and moderation by sex and training time. This study was approved by the institutional ethics committee, and informed consent was obtained from all participants.

Results

Perceived physical, emotional, and social relationship changes were positively associated with self-efficacy; perceived emotional changes correlated strongly with perceived physical changes (r = 0.843, p < 0.01). In structural equation modelling, perceived emotional changes were positioned between perceived physical changes and downstream social relationship changes and self-efficacy, consistent with an indirect association pattern. Sex differences were observed for social relationship changes and self-efficacy, and training time moderated selected paths.

Conclusion

Physical activity–related variables were associated with an interconnected physical–emotional–social pattern linked to self-efficacy, with variation by sex and training time. These findings should be interpreted as model-based associations; longitudinal or experimental studies are needed to establish temporal ordering and causal mechanisms.

Graphical Abstract

Multi-dimensional Pathways of WB Shaped by PA through Exercise Cognition and Training Duration

Supplementary Information

The online version contains supplementary material available at 10.1186/s40359-026-04255-5.

Highlights

This study identifies the multidimensional pathways through which Exercise Cognition Changes and exercise duration are associated with four dimensions of well-being.

Exercise Cognition Changes show significant associations with physical health and emotional well-being.

Average training time emerges as the variable most strongly associated with changes in social relationships.

Physical changes are linked to social relationships indirectly through emotional well-being.

These findings offer useful evidence to support mental health and physical activity programs in higher education settings.T.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40359-026-04255-5.

Introduction

With the rapid pace of social development, the physical and mental health of university students has become an increasing public health concern [1–3]. University students face academic, social, and career-related pressures that make them particularly vulnerable to anxiety, depression, and loneliness, all of which can severely affect their overall well-being (WB). WB is a multidimensional construct encompassing emotional satisfaction, life meaning, physical health, social relationships, and self-efficacy (SE)—that is, an individual’s belief in their ability to achieve specific goals [4].

This study is grounded in Social Cognitive Theory (SCT) and Self-Determination Theory (SDT). SCT highlights reciprocal interactions among person, behavior, and environment, positioning self-efficacy, shaped by exercise mastery experiences, as a key determinant of sustained engagement and positive outcome expectancies [5, 6]. SDT complements this account by proposing that exercise supports well-being insofar as it satisfies autonomy, competence, and relatedness, thereby fostering self-determined motivation and mental health [7]. This study therefore conceptualize physical activity (PA) as being associated with multidimensional well-being via self-efficacy and need satisfaction, and examine the hypothesized structural associations/indirect paths [8, 9].

PA has been widely recognized as an effective intervention for promoting both physical and mental health. Empirical evidence shows that regular PA can alleviate stress and depressive symptoms while improving emotional regulation, SE, and social connectedness [10–14]. Despite these benefits, insufficient PA among young adults has become a global public health issue, with more than 80% of adolescents failing to meet recommended activity levels [15].

Physiologically, PA enhances endocrine balance and metabolic function, thereby promoting emotional changes (EC) through endorphin release [16, 17]. Psychologically, consistent participation in PA strengthens SE by building confidence and persistence [18]. Socially, team- or group-based PA fosters social relationship changes (SRC) by enhancing cooperation, social support, and a sense of belonging [19, 20]. Collectively, these findings suggest that PA promotes WB through interconnected physical, emotional, and social mechanisms.

Despite robust evidence that PA protects university students’ mental health, and emerging work testing mediators such as exercise attitudes, self-control, psychological vulnerability, and resilience [21, 22], several gaps remain. First, theory integration is limited: although WB comprises interrelated domains including affective states, social relationships, SE, and physical health, prior studies often treat WB as a unitary outcome or examine only partial, pairwise associations, leaving unclear how PA may operate through coordinated physiological, psychological, and social processes to promote WB [23]. Second, mechanistic evidence for key indirect pathways remains insufficient, particularly for a sequential body to mind to social pathway. In university students, whether and how affective states bridge PA-related physiological gains such as improved fitness to subsequent improvements in social relationships and SE has not been clearly established, and SEM-based tests of this sequential mediation remain limited [24]. Third, translation remains underdeveloped: interventions tailored to university students and grounded in these complex mechanisms are scarce, and many recommendations do not fully account for contextual features of the transition to adulthood, including heightened stress exposure and lifestyle change, which can reduce specificity and implementability. SCT and SDT offer an integrative framework to address these gaps by specifying how PA may relate to multidimensional WB through SE and basic psychological need satisfaction within a single model.

Accordingly, to address these gaps, the present study used SEM to develop and test a multidimensional integrative model. The model jointly incorporates changes in physical status, affect, social relationships, and SE, and further evaluates exercise cognitions as mediators and gender as a moderator. This approach delineates the pathways and boundary conditions through which PA may contribute to university students’ WB, and provides an empirical basis to inform more targeted health-promotion programmes in higher education.

Materials and methods

Study Design

This study employs a cross-sectional design to explore the relationships between PC, EC, SRC, and SE in university students. Additionally, it aims to identify correlates of these variables, including sex, age, academic background, and PA habits and motivations. Data were collected via a questionnaire survey, systematically recording participants’ demographic characteristics, PA participation, and psychosocial variables. Quantitative analyses were conducted to examine cross-sectional associations and evaluate a hypothesized structural model, without inferring causality.

Participants

The inclusion criteria for this study were as follows: participants had to be full-time university students aged 18 years or older, with at least a bachelor’s degree, regardless of sex. A total of 1,082 university students were included in this study. The majority of participants were male (841, 77.73%), and 241 were female (22.27%). Participants were recruited primarily through two channels. First, an electronic recruitment notice and survey link were distributed via online student communities at multiple Chinese universities, predominantly in Northern China (e.g., departmental WeChat groups and course communication groups). Second, brief on-site study information and recruitment were conducted at the end of selected public elective courses (e.g., physical education and mental health courses). Most participants (89.37%) were between 18 and 22 years old, with a small proportion under 18 years (3.88%) or over 25 years (6.75%). The mean age of male participants was 20.34 ± 1.92 years, and that of female participants was 20.14 ± 1.90 years. In terms of PA participation, participants were required to have engaged in any form of PA during the past 7 days, although no minimum duration of activity was required. Even if a participant’s total activity time was less than one hour in the past week, they were still eligible for inclusion if they expressed a willingness to engage in exercise or had previous exercise experience. Participants also needed to be in good physical health, free from major chronic conditions (e.g., heart disease, diabetes) or recent significant surgeries, and demonstrate basic emotional stability (e.g., no severe depression or anxiety). Additionally, participants had to voluntarily consent to participate in the study and be willing to share their experiences regarding associations of PA on their physical health, emotions, social relationships, and SE.

Exclusion criteria included: 1) individuals with major health issues that would prevent them from engaging in PA or affect their daily lives (e.g., heart disease, recent major surgeries); individuals with severe psychological or emotional issues (e.g., major depression or severe anxiety that significantly impacts daily life), which were preliminarily identified based on self-reported screening questions and World Health Organization Five-Item Well-Being Index (WHO-5) WB Index scores. Such comorbid anxiety-depression conditions are clinically distinct syndromes characterized by altered neurobiological and functional profiles [25], which could confound the psychological and behavioral associations examined in this study; 3) non-voluntary participants, or those who did not provide valid responses during the survey or refused to cooperate. A detailed inclusion and exclusion flowchart is provided in Figure S1. All participants voluntarily consented to the study, and the research was approved by the ethics committee. The detailed inclusion and exclusion criteria are provided in Table S1.

Sample Size

The required sample size was estimated using G*Power 3.1.9.7. Assuming a statistical power of 0.80, a significance level (α) of 0.05, and a medium effect size (f² = 0.15, equivalent to Pearson r ≈ 0.3) according to Cohen (1988), a linear multiple regression model (fixed model, R² deviation from zero) with up to five predictors was specified. The minimum required sample size was calculated to be 176 participants. To ensure sufficient statistical power and account for potential data loss or participant dropout, a total of 1,082 university students were ultimately included in the study, thereby increasing the precision and reliability of the results.

Questionnaire Design

The questionnaire was systematically developed (Supplementary materials) to assess university students’ PA level, SE, health status, and WB. It consisted of four main components: the General Self-Efficacy Scale (GSE) to measure perceived SE; three research-based subscales (PC, EC, and SRC) to capture subjective changes associated with PA; the Global Physical Activity Questionnaire (GPAQ) to assess objective activity level; and two standardized health and well-being measures, the SF-36 and WHO-5, to evaluate overall health and psychological WB.

The GSE [26, 27] assesses individuals’ confidence in their ability to handle challenging tasks and was developed by Schwarzer and colleagues in Germany in 1981 [28]. The scale comprises 10 items rated on a 4-point Likert scale (1, not at all true; 4, exactly true), yielding a total score of 10–40, with higher scores indicating stronger self-efficacy. For interpretability, scores ≥ 30 were described as higher self-efficacy and scores < 30 as lower self-efficacy. In this study, internal consistency was excellent (Cronbach’s α = 0.91), and confirmatory factor analysis supported a single-factor structure with standardized loadings > 0.60, indicating good construct validity.

Three research-based subscales, developed through literature review and pilot interviews, were designed to measure subjective changes resulting from PA—namely PC, EC, and SRC (see the matrix section in the Supplementary materials). PC was defined as perceived positive changes in physical fitness, vitality, and functional status following PA, reflected by improved endurance, increased energy, and reduced daily fatigue [29]. EC was defined as perceived positive changes in affective experience and emotion regulation following PA, reflected by greater positive affect (e.g., enjoyment, calmness) and improved self-regulation under stress or negative mood [30]. SRC was defined as perceived positive changes in interpersonal interaction and social connectedness in exercise contexts or following PA-related psychological improvements, reflected by stronger perceived social support, higher-quality and closer interactions with peers or teammates, and a stronger sense of belonging [31]. Each subscale contained five items rated on a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). All items were positively worded, with no reverse-scored items. Higher scores reflected stronger perceived positive changes in each domain: PC captured perceived improvements in physical strength, endurance, and vitality; EC measured emotional regulation and mood enhancement; and SRC assessed perceived improvements in interpersonal relationships and sense of belonging. In this study, Cronbach’s α coefficients were 0.88 for PC, 0.90 for EC, and 0.87 for SRC. Confirmatory factor analysis supported the hypothesized three-factor model with acceptable fit indices, and indicated good discriminant validity among the three constructs.

The GPAQ, developed by the World Health Organization, provides standardized surveillance of PA across work, transport, leisure, and sedentary domains [32]. GPAQ was used to quantify participants’ PA; for interpretability, an average daily activity time > 60 min was described as higher PA and < 60 min as lower PA, corresponding to a more active versus more sedentary lifestyle.

The 36-Item [33] assesses health-related quality of life and was developed in the United States as part of the Medical Outcomes Study; it is widely used in clinical and public health research. SF-36 comprises eight domains, including physical functioning, pain, vitality, and social functioning [34, 35]. In this study, SF-36 was included as a comprehensive health indicator to capture broad effects of PA on overall physical and psychological functioning rather than isolated emotional or social outcomes. Domains such as vitality, social functioning, and role limitations due to emotional problems conceptually align with PC, EC, and SRC, supporting interpretation of PA effects within an integrated health and psychological well-being framework and enhancing comparability with prior work. In the present sample, Cronbach’s α across the eight domains ranged from 0.78 to 0.92, and construct validity was supported by confirmatory factor analysis.

The WHO-5 [36], developed by the World Health Organization, is a brief, generic screening tool for positive mental health that has been applied worldwide; a validated Chinese version was used to assess positive psychological WB. The WHO-5 comprises five items rated on a 6-point Likert scale (0–5), with a total score ranging from 0 to 25. Higher scores (> 20) indicate greater WB, positive mood, and vitality, whereas lower scores (< 20) suggest reduced WB or potential risk of depressive symptoms. In this sample, internal consistency was good (Cronbach’s α = 0.86).

All items across scales were positively scored, with no reverse-coded items. Each subscale score represented the mean of its items (five for PC, EC, and SRC; ten for GSE; detailed items are provided in the Supplementary materials). The overall questionnaire score was calculated as the average of the four subscale means, where higher scores indicated greater positive experiences, SE, and WB associated with PA.

Statistical Analysis

SPSS 19.0 and AMOS 21.0 were used for data processing and analyses to evaluate the measurement quality of the scales, support the validity of the study-developed instruments, and verify model validity for subsequent specification. Cronbach’s α values were calculated using SPSS 19.0 to assess the internal consistency of the questionnaire scales. Confirmatory factor analysis (CFA) was conducted using AMOS 21.0 to assess the construct validity of the survey scales. To verify the independence of the latent variables, three models were constructed: a three-factor model, a two-factor model (e.g., combining PC and EC), and a one-factor model (combining all latent variables). Model fit was assessed using indices such as Chi-Square Min/Degrees of Freedom (CMIN/DF), Root Mean Square Error of Approximation (RMSEA), Comparative Fit Index (CFI), and Tucker-Lewis Index (TLI).

Before conducting confirmatory analyses, exploratory factor analysis (EFA) was performed to preliminarily examine the latent factor structure of the study-developed scales. EFA used varimax rotation to extract factors and to evaluate the adequacy of factor loadings. To address research questions on group differences, independent-samples t tests and analysis of variance (ANOVA) were used to assess the effects of grouping variables (e.g., gender and interest in exercise) on outcome variables. Correlation analyses were conducted to examine linear associations among latent constructs.

To systematically test a hypothesized SEM of associations and indirect paths (product-of-coefficients), interpreted as statistical indirect associations given cross-sectional data involving PC, EC, and SRC. SEM was conducted using AMOS 21.0 to evaluate the relationships and impact pathways between latent variables. To test the statistical significance of the indirect effects/indirect associations, a bootstrapping procedure with 5,000 resamples was employed using the bias-corrected percentile method in AMOS 21.0. Indirect effects were considered significant if the 95% confidence intervals did not include zero, providing a robust evaluation of the mediational pathways. Bootstrapping was applied to calculate the standard errors and significance levels of the parameters, ensuring the robustness of the model results. Different methods were chosen based on the variable types: Pearson correlation was used for relationships between continuous variables, while Spearman correlation was applied for relationships involving non-normally distributed or categorical variables. The correlation matrix was visualized using R software with the ggplot2 package, which generated heatmaps to intuitively display the correlation strength and statistical significance between variables through color gradients and significance markers.

To ensure the rigor of the scale’s measurement structure, the study conducted item-level examinations of factor loadings and cross-loadings for all measurement items of PC, EC, and SRC following the confirmatory factor analysis. This allowed a detailed evaluation of the item-level structure of the measurement model. In addition, multigroup structural equation modeling (MG-SEM) was employed to assess measurement invariance across sex, with configural, metric, and scalar invariance models tested sequentially.

Results

Demographics and Characteristics of Participants

A total of 1,082 university students participated in this study (Table 1). The sample was predominantly male, and most participants were aged 18–22 years. Cross-tabulation analyses indicated significant associations between gender and both sports interest and weekly activity duration (χ² tests, p < 0.001), with higher PA levels and stronger exercise interest more frequently reported among male students (Table S2).

Table 1.

Basic Information of the Participants (N = 1082)

Variables	Category	N (%)
Sex	Male	841 (77.73%)
	Female	241 (22.27%)
Age range	below 18	42 (3.88%)
	18–22	967 (89.37%)
	23–25	60 (5.55%)
	26–30	5 (0.46%)
	above 30	8 (0.74%)
Academic status	Undergraduate	837 (77.36%)
	Master	20 (1.85%)
	Doctoral	12 (1.11%)
	Other	213 (19.69%)
Major	Humanities	38 (3.51%)
	Social studies	16 (1.48%)
	Natural sciences	12 (1.11%)
	Science and Engineering	834 (77.08%)
	Biomedical Sciences	2 (0.18%)
	Management discipline	24 (2.22%)
	Art subjects	46 (4.25%)
	Physical Education	5 (0.46%)
	Other	105 (9.7%)
Interest in sports	Yes	800 (73.94%)
	No	282 (26.06%)
Weekly physical activity duration	0–1 h	358 (33.09%)
	1–3 h	369 (34.1%)
	4–8 h	216 (19.96%)
	9–15 h	90 (8.32%)
	more than 15 h	49 (4.53%)
Motivation for sports activity (multiple choice)	Long-term exercise habits	533 (49.26%)
	The desire to keep healthy	477 (44.09%)
	Unhealthy appearance	292 (26.99%)
	Improve body shape	441 (40.76%)
	Mental condition	337 (31.15%)
	Emotional status	218 (20.15%)
	Be influenced by others	186 (17.19%)
Changes experienced through sports (multiple choice)	Health	799 (73.84%)
	Endurance	640 (59.15%)
	Stress Relief	581 (53.7%)
	Emotional Expression	469 (43.35%)
	Social Relationships	254 (23.48%)
	Self-Awareness	311 (28.74%)
	Aesthetic Awareness	225 (20.79%)
	Others	31 (2.87%)
Experience of feeling incompetent or insecure in sports or exercise	Yes	400 (36.97%)
	No	682 (63.03%)
Having a sense of accomplishment in sports will make you more confident in dealing with other problems	Yes	607 (56.1%)
	No	475 (43.9%)

Most participants were undergraduates and were enrolled in science and engineering programmes. Interest in PA was common (73.94%). Weekly Exercise Duration (ED) was concentrated at the lower-to-moderate end, with approximately one-third reporting < 1 h per week (33.09%) and a further one-third reporting 1–3 h per week (34.1%), whereas 12.85% reported > 9 h per week.

Motivations for engaging in PA were heterogeneous, most commonly long-term exercise habits (49.26%) and health maintenance (44.09%). Reported benefits were also heterogeneous, most frequently perceived health improvement (73.84%) and stress reduction (53.7%), and 28.74% reported increased self-awareness.

During PA, feelings of incompetence or low confidence were reported by 36.97% of participants. Conversely, 56.1% agreed that achievement during PA increased their confidence in addressing other challenges.

Reliability and Validity Verification (VV) Supporting the Reliability and Scientific Rigor of the Research Model

Reliability and validity analyses were conducted to evaluate measurement quality and structural validity (Table S3–S6). CFA supported the measurement model, with strong standardized loadings and evidence of convergent validity based on AVE and CR (Table S3–S6). Overall model fit was acceptable (χ²/df = 4.67, CFI = 0.955, TLI = 0.946, RMSEA = 0.093). Discriminant validity was also supported, as the hypothesized three-factor structure (PC/EC/SRC) fit better than alternative models (Table S6), indicating separation among the latent constructs.

Based on the acceptable CFA results, item-level structure and cross-sex measurement stability were further examined. Item-level diagnostics showed adequate factor structure, with high primary loadings and limited cross-loadings (Table S7). Multigroup measurement invariance testing indicated that configural, metric, scalar, and strict invariance were all supported across sex groups (ΔCFI < 0.01; Table S8), suggesting comparable measurement properties between males and females. No items were removed. Internal consistency was high across all dimensions (Cronbach’s α > 0.90; PC = 0.923, EC = 0.959, SRC = 0.943, SE = 0.981), supporting scale reliability.

Significant Positive Correlations Among Core Variables Reveal the Intrinsic Synergy of Individual Changes

Pearson correlation analyses were conducted to examine associations among PC (A), EC (B), SRC (C), and SE (D) (Table S9). All four variables were positively correlated (p < 0.01). The strongest correlations were observed between PC and EC (r = 0.843) and between EC and SRC (r = 0.821), followed by PC and SRC (r = 0.781) and SRC and SE (r = 0.749) (Table S9; Fig. 1). Together, these results indicate consistent positive covariation among PC, EC, SRC, and SE, supporting subsequent pathway modelling.

Fig. 1.

Significant Positive Correlations Between Physical Health, Emotion, Social Relationships, and SE (Pearson Correlation Analysis)

Influence of sex Differences on Core Variables

Spearman correlation analyses were conducted to examine associations of age, sex, academic degree, and major with PC, EC, SRC, and SE (Fig. 2). Overall, age, academic degree, and major showed weak correlations with the core variables. In contrast, sex showed clearer associations with SRC and SE. Consistent with this pattern, Table 2 indicated no sex differences in PC or EC (t = 0.672, p = 0.502; t = 0.954, p = 0.34, respectively). However, males reported higher SRC (t = 3.216, p = 0.001) and SE (t = 2.185, p = 0.029) than females.In contrast, correlations of academic degree, age, and field of study with the core variables were weak (ρ = 0.057).

Fig. 2.

Effects of Sex, Age, and Education Level on Core Variables (Physical Health, Emotional WB, Social Relationships, and SE)

Table 2.

The Relationship between Physical Changes, Emotional Changes, Social Relationship Changes, Self-efficacy and Sex

Index	Sex	N	Mean	SD	t	P
Physical Changes	Male	841	2.442	0.524	0.672	0.502
	Female	241	2.417	0.503
Emotional Changes	Male	841	2.48	0.537	0.954	0.34
	Female	241	2.442	0.561
Social Relationship Changes	Male	841	2.473	0.548	3.216	0.001
	Female	241	2.344	0.549
Self-Efficacy	Male	841	3.789	0.925	2.185	0.029
	Female	241	3.645	0.889

Correlation Analysis Between PA Habits and Key Variables

To examine associations between PA habits and the core variables, Spearman correlations were calculated for exercise preference, PA time in the past 7 days, and participation in external training with PC, EC, SRC, and SE (Fig. 3). Exercise preference was positively correlated with EC (ρ = 0.266, p < 0.01) and SE (ρ = 0.256, p < 0.01). PA duration was positively correlated with EC (ρ = 0.161, p < 0.01), SRC (ρ = 0.190, p < 0.01), and SE (ρ = 0.205, p < 0.01). Participation in external training was positively correlated with EC (ρ = 0.227, p < 0.01) and SE (ρ = 0.247, p < 0.01). Correlations for specific activity types are shown in Fig. 3.

Fig. 3.

Correlations Between PA Habits and Core Variables, Including Exercise Preferences and Duration

In line with these associations, Table 3 showed higher PC, EC, SRC, and SE scores among participants reporting strong exercise interest than among those reporting lower interest (all p < 0.01): PC 2.51 vs. 2.23 (t = 8.011), EC 2.561 vs. 2.219 (t = 9.457), SRC 2.52 vs. 2.26 (t = 10.556), and SE 3.79 vs. 3.48 (t = 8.81) (Table 3).

Table 3.

The Relationship between Physical Changes, Emotional Changes, Social Relationship Changes, Self-efficacy and Interest in Sports

Index	Interest in sports	N	M	SD	t	P
Physical Changes	Yes	800	2.51	0.487	8.011	0
	No	282	2.23	0.551
Emotional Changes	Yes	800	2.561	0.5	9.457	0
	No	282	2.219	0.578
Social Relationship Changes	Yes	800	2.545	0.503	10.556	0
	No	282	2.161	0.582
Self-Efficacy	Yes	800	3.901	0.875	8.81	0
	No	282	3.348	0.918

Analysis of the Relationship Between Exercise Type, Training Duration, and Core Variables

Associations of average training duration and sustained activity types with the core variables were examined using Spearman correlations (Fig. 4). Average training duration was positively correlated with the core variables, with the largest coefficients observed for SRC (ρ = 0.200, p < 0.01) and SE (ρ = 0.197, p < 0.01).

Fig. 4.

Effects of ATT and Sustained Exercise Types on Core Variables (Physical, Emotional, Social Relationships, and SE)

Activity-specific correlations showed heterogeneity across sustained exercise types (Fig. 4). Positive correlations were observed for several commonly reported activities, whereas some activity categories showed coefficients close to zero. “Other” sustained activities showed negative correlations with EC (ρ = −0.063, p < 0.05) and SRC (ρ = −0.070, p < 0.01). Overall, these results indicate that both training duration and activity type are associated with variation in the core variable.

Associations Between Different Exercise Environments on PC, EC, SRC, and SE

Associations between exercise environment types and PC, EC, SRC, and SE were examined using correlation analyses (Fig. 5). Team-based environments were positively correlated with all four variables, with larger coefficients observed for SRC (ρ = 0.186, p < 0.01) and EC (ρ = 0.118, p < 0.01). Competitive environments also showed positive correlations with EC (ρ = 0.053, p < 0.05) and SRC (ρ = 0.105, p < 0.01). In contrast, traditional exercise environments showed correlations close to zero across PC, EC, SRC, and SE (Fig. 5).

Fig. 5.

Impact of Different Exercise Environments on PC, EC, Social Relationships, and SE

Relaxation-based activities (e.g., yoga, meditation) and creative exercise (e.g., dance) showed no significant correlations with the core variables (Fig. 5). Solitary activities were weakly correlated with the core variables and showed negative correlations with EC and SE (Fig. 5). Overall, correlation patterns differed by exercise environment type, with positive associations concentrated in team-based and competitive environments.

Analysis of Relationship Between Exercise Goals on the Dimensions of WB

This study examined associations between exercise objectives and PC (A), EC (B), SRC (C), and SE (D) (Fig. 6). Long-term exercise habits were positively correlated with all four dimensions, with higher coefficients observed for EC (ρ = 0.182, p < 0.01) and SE (ρ = 0.221, p < 0.01). Physical fitness–oriented goals were also positively correlated with the physical, emotional, and social dimensions, with small effect sizes (ρ = 0.061–0.094, p < 0.05). Appearance-oriented goals showed positive correlations with EC and SRC, whereas stress-relief goals were positively correlated with EC and SRC (Fig. 6). In contrast, environmental influences were negatively correlated with all dimensions, with larger coefficients for PC (ρ = −0.102, p < 0.01) and SE (ρ = −0.103, p < 0.01). Overall, correlation patterns differed across exercise objectives.

Fig. 6.

Analysis of Exercise Goals on Physical, Emotional, Social Relationships, and SE

Associations Between Exercise and Changes in Cognition, Physical/Emotional/Social Relationships, and Self-Efficacy

This section examined associations between dimensions of Exercise Cognition Changes (ECC) and PC (A), EC (B), SRC (C), and SE (D) (Fig. 7). Perceived importance of exercise showed positive correlations with all four outcomes (ρ = 0.303–0.328 across PC, EC, SRC, and SE). Age-related ECC factors were also positively correlated with all outcomes, with relatively larger coefficients for SRC and SE. Health awareness and perceived physical condition showed small positive correlations, most consistently with EC and SRC. Environmental ECC factors were positively correlated with EC, SRC, and SE, with coefficients concentrated in EC and SRC. Overall, ECC dimensions showed broadly positive correlation patterns across the four outcomes.

Fig. 7.

Comprehensive Analysis of Exercise Cognition Changes on Physical, Emotional, Social Relationships, and SE

Associations of Exercise Cognition and Average Training Time (ATT) With Physical, Emotional, Social, and Self-Efficacy Dimensions

Model fit was acceptable in both SEMs (Tables 4 and 5) [37]. The χ²/df values were 4.673 and 4.155 (both < 5), and complementary fit indices (GFI, CFI, NFI, TLI, RMSEA, RMR) met conventional criteria (Fig. 8A and B).

Table 4.

Analysis of the Overall Goodness of Fit of the Model of Cognition Change Importance on Physical, Emotional, Social, and Self-Efficacy

Fit Index	Acceptance Criteria	Model Result	Conclusion
χ² (Chi-square)	-	467.3	-
df (Degrees of Freedom)	-	100	-
χ²/df (CMIN/DF)	< 5	4.673	Acceptable
GFI	> 0.9	0.916	Good
CFI	> 0.9	0.972	Good
NFI	> 0.9	0.964	Good
TLI	> 0.9	0.967	Good
RMSEA	< 0.08	0.058	Good
RMR	< 0.05	0.012	Good

CMIN/DF Chi-Square Min/Degrees of Freedom, RMR Root Mean Square Residual, GFI Goodness of Fit Index, IFI Incremental Fit Index, TLI Tucker-Lewis Index, CFI Comparative Fit Index, RMSEA Root Mean Square Error of Approximation

Table 5.

Analysis of the Overall Goodness of Fit of the Model of Average Training Time on Physical, Emotional, Social, and Self-Efficacy

Fit Index	Acceptance Criteria	Model Result	Conclusion
χ² (Chi-square)	-	415.5	-
df (Degrees of Freedom)	-	100	-
χ²/df (CMIN/DF)	< 5	4.155	Acceptable
GFI	> 0.9	0.928	Good
CFI	> 0.9	0.977	Good
NFI	> 0.9	0.97	Good
TLI	> 0.9	0.972	Good
RMSEA	< 0.08	0.054	Good
RMR	< 0.05	0.011	Good

CMIN/DF Chi-Square Min/Degrees of Freedom, RMR Root Mean Square Residual, GFI Goodness of Fit Index, IFI Incremental Fit Index, TLI Tucker-Lewis Index, CFI Comparative Fit Index, RMSEA Root Mean Square Error of Approximation

Fig. 8.

SEM Analysis.

In the ECC model, ECC showed positive associations with PC, EC, and SE, with the strongest path observed for the path from ECC to PC (β = 0.349, p < 0.001) (Table 6). EC and SRC were positively associated with PC. No directpath from ECC to SRC path was observed; instead, ECC showed indirect associations with SRC via PC and EC (Table 6). PC and SRC were each positively associated with SE (Table 6).

Table 6.

Path Relationship between Cognition Change Importance, Physical, Emotional, Social, and Self-Efficacy

Path Relationship	Standard Path Coefficient	S.E.	C.R.	P
CCI→PC	0.349	0.019	11.122	***
CCI→EC	0.042	0.012	2.364	0.018
PC→EC	0.894	0.038	26.943	***
EC→SRC	0.542	0.055	9.971	***
PC→SRC	0.381	0.063	6.93	***
CCI→SE	0.045	0.026	2.187	0.029
PC→SE	0.268	0.11	5.009	***
SRC→SE	0.547	0.098	10.065	***

PC Physical Changes, EC Emotional Changes, SRC Social Relationship Changes, CCI Cognition Change Importance, SE Self-Efficacy

In the ATT model, ATT was positively associated with PC (β = 0.174, p < 0.01) and SRC (β = 0.436, p < 0.001), whereas no direct association between ATT and EC was observed (Table 7). An indirect association between ATT and EC was observed via PC (Table 7). EC and PC were positively associated with SRC (β = 0.478, p < 0.001), and both PC and SRC were positively associated with SE (β = 0.285 and β = 0.54, respectively) (Table 7; Fig. 8B).

Table 7.

Path Relationship between Path Relationship Between Average Training Time, Physical, Emotional, Social, and Self-Efficacy, Physical, Emotional, Social, and Self-Efficacy

Path Relationship	Standard Path Coefficient	S.E.	C.R.	P
ATT→PC	0.174	0.013	5.519	***
PC→EC	0.923	0.038	27.255	***
EC→SRC	0.478	0.067	7.251	***
PC→SRC	0.436	0.078	6.409	***
ATT→SRC	0.04	0.008	2.307	0.021
ATT→SE	0.047	0.016	2.378	0.017
PC→SE	0.285	0.118	4.907	***
SRC→SE	0.54	0.105	9.106	***

PC Physical Changes, EC Emotional Changes, SRC Social Relationship Changes, SE Self-Efficacy, ATT Average Training Time

Across both models, SE was associated with ECC- and ATT-linked pathways spanning physical, emotional, and social constructs (Fig. 8A and B; Tables 6 and 7). EC showed associations with PC and with downstream links to SRC and SE (Tables 6 and 7), while SRC was primarily linked through paths involving PC and EC; ATT showed a direct association with SRC (Fig. 8B).

Model explanatory performance was evaluated using R², f², and bootstrapped indirect effects. R² values for PC, EC, SRC, and SE were 0.58, 0.67, 0.63, and 0.71, respectively. Effect sizes (f²) for main structural paths ranged from 0.15 to 0.37. Bootstrapping (5,000 resamples) indicated significant indirect effects, with 95% confidence intervals excluding zero (Table S10). Overall, the models showed acceptable fit and stable structural associations (Tables 4, 5, 6 and 7; Fig. 8A and B). Overall, the model exhibited good fit and stable structural paths.

Discussion

With rapid social development, university students’ physical and mental health has become an increasing concern. PA, as an integrated health-promoting behaviour, has been linked to physical health, psychological regulation, and SE. However, prior research has often focused on single psychological variables or overall WB, and systematic characterisation of multidimensional pathways linking PA to WB remains limited [38–40]. Using SEM, this study examined the structural associations among PC, EC, SRC, and SE within a unified framework, and further tested moderation by ATT and demographic factors.

PA was positively associated with PC, EC, SRC, and SE. Associations were comparatively stronger for PC with EC, EC with SRC, and SRC with SE [41, 42], indicating coordinated variation across physical, emotional, and social domains. EC showed the strongest indirect pathway (indirect association) between PC and downstream SRC/SE in the hypothesized model (Fig. 9). This pattern aligns with evidence implicating psychological and social factors in WB formation, including pathways such as “self-identity–self-esteem–SE” [43, 44] and “perceived stress–peer relationships” [45, 46].

Fig. 9.

Mediating role of emotional changes and moderating effects of sex and exercise duration

Sex and ATT showed moderation effects in the model. Male participants reported higher mean levels of SRC and SE than female participants, suggesting that differences may relate to PA participation contexts and access to structural opportunities. Prior work indicates that gender socialisation may shape preferences for team-based or competitive activities, thereby producing differences in social interaction structures and achievement experiences [47]. Associations between team contexts and EC and SE may also operate through sex-differentiated socio-psychological processes [48, 49]. Accordingly, the concentration of associations in team and competitive contexts supports greater attention to the social-structural features of activity settings (e.g., opportunities for cooperation, shared goals, and interaction density), rather than attributing differences to sex alone.

This study integrates and refines prior evidence at the structural level. Previous work has linked regular physical activity to positive affect and psychological well-being [50] and suggested that exercise-related cognitions and attitudes may influence emotion and mental health via motivational processes [51]. Building on these findings, the present study incorporated ECC into a single structural equation modelling framework and jointly tested perceived physical changes, perceived emotional changes, perceived social relationship changes, and self-efficacy, extending research that has typically focused on a single psychological factor or overall well-being [52, 53]. The model supports the positioning of perceived emotional changes between perceived physical changes and downstream social relationship changes and self-efficacy, providing structured evidence for a progressive physical–emotional–social linkage and strengthening the evidence base for emotion as a bridging component. ECC was specified within the hypothesised pathway and was not evaluated against alternative mechanisms or competing models; accordingly, interpretations are limited to support for the proposed pathway rather than identifying ECC as a unique or key mediator.

At the practical level, these findings provide a reference for university health promotion. Given the sample characteristics (predominantly male and STEM students), PA promotion programmes for similar populations may prioritise ECC-related elements and sustained participation (ATT), and incorporate activity-setting features that support SRC and EC (e.g., cooperative and interactive opportunities) to facilitate improvements in SE and WB-related dimensions. Implementation should consider resource constraints and campus conditions. Small-scale pilots in relatively resource-sufficient departments or student organisations may be used to assess time demands, facilities, staffing, and budget needs. Existing curricular resources (e.g., physical education and mental health education) and digital platforms may also be leveraged to increase flexibility and reduce reliance on fixed venues and concentrated schedules.

Several limitations should be noted. First, the cross-sectional design precludes causal inference; SEM results support statistical associations under the hypothesised model and cannot be equated with temporal causal effects [54]. Second, sampling imbalance (predominantly male, STEM, and 18–22-year-old participants) limits external validity, and replication is needed in more balanced samples across sex, age, and disciplines. Third, self-report measures may be subject to social desirability and recall bias; high Cronbach’s α values support internal consistency but may also indicate item homogeneity and potential redundancy, affecting scale efficiency and conceptual discrimination. Fourth, RMSEA values for some CFA dimensions exceeded the stringent 0.08 threshold, suggesting item redundancy or local misfit and limiting measurement precision. Fifth, higher R² values and strong factor loadings support explanatory performance, but model complexity relative to the sample warrants attention to potential SEM overfitting; external replication and stricter model validation are needed to assess robustness. Sixth, weak or negative associations for dance and yoga may reflect activity intensity, group differences, or self-selection bias; the Chinese university context may also constrain generalisability of PA–WB associations.

Future work should further specify and measure key mechanism variables within Social Cognitive Theory and Self-Determination Theory, employ longitudinal or experimental designs, broaden sample diversity, and integrate objective assessments and multi-source data to test temporal ordering and cross-context applicability [55].

Conclusion

Across 1,082 university students, PA was positively associated with PC, EC, SRC and SE. ECC was associated with PC and EC and related to SRC and SE largely via indirect pathways, whereas ATT was associated with PC and SRC. In the SEMs, EC was positioned between PC and downstream SRC and SE, outlining a multidimensional pattern linking PA-related exposure with physical, emotional and social domains of WB (Graphic abstract). Given the cross-sectional, self-reported design and the imbalanced sample composition, these pathways should be interpreted as model-based associations rather than confirmed mechanisms, and require validation in longitudinal or experimental studies.

Abbreviations

ANOVA

Analysis of Variance

ATT

Average Training Time

AVE

Average Variance Extracted

CFI

Comparative Fit Index

CMIN/DF

Chi-Square Min/Degrees of Freedom

CR

Composite Reliability

EC

Emotional Change

ECC

Exercise Cognition Changes

ED

Exercise Duration

GFI

Goodness of Fit Index

GPAQ

Global Physical Activity Questionnaire

GSE

General Self-Efficacy Scale

PA

Physical Activity

PC

Physical Changes

RMR

Root Mean Square Residual

RMSEA

Root Mean Square Error of Approximation

SE

Self-Efficacy

SEM

Structural Equation Modeling

SF-36

36-Item Short Form Health Survey

SRC

Social Relationship Changes

TLI

Tucker-Lewis Index

WB

Well-being

VV

Validity Verification

WHO-5

World Health Organization-Five Well-Being Index

Authors’ contributions

Xiaomei Zhang and Yaming Wei contributed equally to this work and are regarded as co-first authors. Xiaomei Zhang was responsible for data collection, statistical analysis, and manuscript drafting. Yaming Wei contributed to the study design, interpretation of results, and critical revision of the manuscript. Yongfan Song conceived and supervised the study, secured funding, and provided final approval of the version to be published. All authors read and approved the final manuscript.

Funding

None.

Data availability

All data can be provided as needed.

Declarations

Ethics approval and consent to participate

This study was approved by the Clinical Ethics Committee of Fuzhou University and was conducted in accordance with the Declaration of Helsinki. This study was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki. The research protocol was reviewed and approved by the Clinical Ethics Committee of Fuzhou University. The study posed minimal risk to participants, and all data were collected anonymously to protect participant privacy. All participants provided informed consent prior to their inclusion in the study. They were informed of the study’s purpose, procedures, and their right to withdraw at any time without consequence. Consent was obtained voluntarily from each participant, and no incentives were provided that could influence participation.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.