Social anxiety, physical exercise, regulatory emotional self-efficacy, and social avoidance among junior high school students: a cross-sectional study

Abstract

Objective

This study examined the cross-sectional associations of physical exercise (PE), regulatory emotional self-efficacy (RESE), and social avoidance (SAV) with social anxiety (SA) among junior high school students.

Methods

A school-based cross-sectional survey was conducted among 817 students from five junior high schools in Nanyang, Henan Province, China. PE, RESE, SAV, and SA were assessed using the Physical Activity Rating Scale-3, Regulatory Emotional Self-Efficacy Scale, SAV dimension of the Social Avoidance and Distress Scale, and Social Anxiety Scale for Adolescents, respectively. Data were analyzed using descriptive statistics, zero-order correlations, two-step hierarchical multiple linear regression with HC3 standard errors, and robustness and supplementary analyses.

Results

SA was negatively correlated with PE (r = − 0.320, p < 0.001) and RESE (r = − 0.330, p < 0.001), and positively correlated with SAV (r = 0.434, p < 0.001). After adjustment for sex, grade, and household registration, PE (β = −0.174, p < 0.001) and RESE (β = −0.182, p < 0.001) were negatively associated with SA, whereas SAV was positively associated with SA (β = 0.322, p < 0.001). The final model explained 26.6% of the variance in SA. The directions of the main associations were unchanged after log-transforming PE and excluding potentially influential observations; SAV also remained associated with fear of negative evaluation in the alternative-outcome analysis. Efficacy in managing despondency or distress and efficacy in managing anger or irritation were negatively associated with SA, whereas efficacy in expressing positive affect was not independently significant.

Conclusion

SA among junior high school students appears to be associated with multiple behavioral and psychological domains rather than a single factor, encompassing PE, RESE, and SAV. These findings highlight the importance of differentiated student assessment and support that considers PE participation, confidence in managing difficult emotions, and SAV-related behavior alongside SA symptoms.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s40359-026-05742-5.

Introduction

Social anxiety (SA) is a prominent mental health concern during adolescence, characterized by persistent fear of negative evaluation and heightened tension, distress, or avoidance in social situations [1]. A recent global meta-analysis estimated that social anxiety disorder affects approximately 8.3% of adolescents, underscoring the clinical and public health relevance of severe social anxiety during this developmental period [2]. Adolescence is marked by increasing reliance on peer relationships and heightened sensitivity to social evaluation, making interpersonal contexts central to the development of belonging, self-concept, and social identity [3–6]. During the transition to junior high school, these developmental changes may make ordinary social encounters feel increasingly evaluative and threatening [4, 7]. Higher levels of SA have been associated with poorer peer functioning, reduced social competence, and difficulties in academic and school adjustment [7–9]. Accordingly, identifying behavioral and psychological characteristics associated with SA may help clarify individual differences in social-emotional adjustment among junior high school students.

Among the behavioral factors associated with adolescent SA, physical exercise (PE) represents a measurable and potentially modifiable form of everyday activity. Reviews and meta-analyses have generally associated higher levels of physical activity with more favorable mental health outcomes and lower anxiety among children and adolescents [10–13]. A recent systematic review and meta-analysis of typically developing children and adolescents also found that physical activity interventions were associated with improvements in anxiety and broader mental health outcomes, although the overall certainty of the evidence was rated as low because of methodological limitations in the included studies [14]. More specifically, a systematic review and meta-analysis found that higher physical activity was associated with lower SA symptoms across clinical and non-clinical populations [15]. Adolescent cross-sectional research has similarly reported a negative association between PE and SA [16]. Longitudinal and cross-lagged studies further suggest that PE and SA may be prospectively related in both directions, although these findings do not establish a simple causal pathway [17, 18]. Taken together, the available evidence supports examining PE as a behavioral correlate of adolescent SA. However, the PARS-3 used in the present study assessed overall exercise volume based on intensity, duration, and frequency; it did not distinguish school-based from non-school activity, individual from group exercise, or the social characteristics of participation. On the basis of this evidence, higher PE was expected to be associated with lower SA.

Regulatory emotional self-efficacy (RESE) refers to individuals’ beliefs in their capability to manage negative emotions and express positive emotions, including perceived efficacy in managing despondency or distress, managing anger or irritation, and expressing positive affect [19]. Within social cognitive theory, self-efficacy refers to individuals’ beliefs about their capability to organize and perform actions required to manage particular demands [20]. Emotion-regulation frameworks further emphasize that the ways in which emotions are understood, managed, and expressed are associated with psychological adjustment, but confidence in regulating emotions should be distinguished from the frequency or effectiveness of specific regulation strategies [21–23]. Socially evaluative situations may evoke tension, shame, worry, or embarrassment among adolescents, making perceived confidence in managing these emotions potentially relevant to SA. However, the present cross-sectional framework examines only whether RESE is associated with SA and does not test whether RESE changes threat appraisals, emotional responses, or subsequent coping behavior. Studies conducted in university, high-school, and junior-high-school samples have associated RESE with anxiety-related symptoms, negative emotions, or interpersonal adjustment, although much of this evidence comes from populations or outcomes that differ from those examined in the present study [24–27]. Clinical studies have reported emotion-regulation difficulties and less adaptive patterns of strategy use among individuals with social anxiety disorder, but these findings concern clinical populations and do not necessarily represent RESE as measured in general junior high school students [28, 29]. Recent psychometric evidence from a large Chinese sample suggests that the RESE scale contains a dominant general factor together with meaningful domain-specific components, providing support for examining an overall RESE score while also considering its specific dimensions [30]. On the basis of this evidence, higher RESE was expected to be associated with lower SA.

Social avoidance (SAV) refers primarily to a behavioral tendency to reduce, withdraw from, or avoid social contact in situations perceived as uncomfortable or threatening [31]. SAV is therefore conceptually narrower than SA, which is a broader symptom construct encompassing fear of negative evaluation as well as avoidance and distress in new and general social situations [32]. Although SAV and SA are not interchangeable, they are not fully independent because avoidance-related content is represented in both the SAV measure and two dimensions of the SAS-A used to assess SA. Cognitive-behavioral accounts of SA identify avoidance and safety behaviors as closely related behavioral features of socially anxious functioning [33]. Studies involving adolescents with social anxiety disorder have reported elevated negative social cognitions and greater use of avoidant safety behaviors, whereas broader adolescent research has also linked avoidance-related behavior with higher SA symptoms [34, 35]. Avoidance has been theorized to limit exposure to social situations in which negative expectations might otherwise be tested, but the present study did not measure social exposure, peer feedback, or changes in social expectations [33]. Longitudinal research has identified distinct trajectories of social avoidance across adolescence and found that higher social anxiety was associated with an increasing avoidance trajectory [36]. Other longitudinal evidence has shown that fear of negative evaluation and social avoidance are positively but only weakly related and may display different associations with peer status, further supporting their treatment as related but distinguishable features [37]. Taken together, the available evidence supports examining SAV as a behavioral correlate of adolescent SA, while its conceptual and measurement proximity to the avoidance-related dimensions of SA requires cautious interpretation. For this reason, the present study considered the association between SAV and total SA alongside a supplementary analysis using fear of negative evaluation, which does not contain avoidance items, as an alternative outcome. On the basis of this evidence, higher SAV was expected to be associated with higher SA.

Taken together, PE, RESE, and SAV represent three conceptually distinct domains that may be associated with adolescent SA: everyday exercise behavior, beliefs about emotion-regulation capability, and avoidance tendencies in social situations, respectively. Previous studies have examined several of these relationships separately or in partial combinations, including the association between PE and SA, associations between RESE and anxiety-related outcomes, and relationships among physical activity, SAV, and SA [24, 38, 39]. In the present study, these variables were treated as parallel correlates of SA rather than as components of a developmental sequence, a mediation model, or a causal pathway. However, comparatively little research has simultaneously estimated the independent associations of PE, overall RESE, and SAV with SA in the same general sample of junior high school students. Accordingly, this school-based cross-sectional study first described the zero-order associations among PE, RESE, SAV, and SA and then examined whether PE, RESE, and SAV remained independently associated with SA after adjustment for sex, grade, and household registration. On the basis of the preceding evidence, the following hypotheses were proposed:

H1

Higher PE would be associated with lower SA.

H2

Higher RESE would be associated with lower SA.

H3

Higher SAV would be associated with higher SA.

Sex-by-predictor interactions and grade differences in SA were examined as exploratory analyses rather than as primary hypotheses.

Methods

Participants and study design

Junior high school corresponds to early adolescence, a developmental period during which experiences related to SA are likely to emerge. After entering junior high school, students typically face increasing peer salience, heightened sensitivity to others’ evaluations, and physiological and psychological changes associated with puberty, all of which may increase evaluative pressure and adjustment demands in social situations [40–42]. Therefore, this study recruited junior high school students and adopted a cross-sectional questionnaire design. Data were collected from March 18 to March 25, 2026, using convenience sampling in five junior high schools in Nanyang, Henan Province, China: Nanyang No. 3 Junior High School, Nanyang No. 9 Junior High School, Nanyang No. 13 Junior High School, Nanyang No. 21 Junior High School, and Nanyang No. 22 Junior High School. Within each school, participating classes from Grades 7, 8, and 9 were identified with the assistance of school administrators. A total of 16 classes participated in the survey. One class was included from each grade in each school, except for Grade 8 at Nanyang No. 3 Junior High School, where two classes participated. Homeroom teachers distributed the online questionnaire link to students through class group chats, and students completed the questionnaire independently online. This was a school-based cross-sectional survey in the sense that participants were recruited through schools and accessed the questionnaire through class-based online distribution; the instruments assessed general PE, RESE, SAV, and SA rather than school-context-specific constructs.

A total of 932 questionnaires were distributed and 871 were returned, corresponding to a return rate of 93.45%. After screening, 54 questionnaires were excluded: 13 solely because the completion time was less than 180 s, 30 solely because the same response option was selected throughout the questionnaire, and 11 because they met both criteria. The final analytic sample comprised 817 valid questionnaires, representing 93.80% of the returned questionnaires and 87.66% of all distributed questionnaires. The five schools contributed 163, 142, 168, 175, and 169 valid questionnaires, respectively; the school- and grade-specific distributions are provided in Supplementary Table S1.

To improve coverage across grade levels, the survey was administered to classes from all three grades in each participating school. The sample included students of different sexes, grades, and household registration statuses, providing a basis for adjustment for basic demographic variables and supplementary analyses. The inclusion criteria were as follows: (1) students who voluntarily participated after written informed consent had been obtained from both the students and their parents or legal guardians; and (2) students currently enrolled in one of the participating schools. This study was conducted in accordance with the principles of the Declaration of Helsinki. Written informed consent was obtained from all participating students and their parents or legal guardians. The study protocol was approved by the Ethics Committee of the Academic Committee of Capital University of Physical Education and Sports (approval number: 2026A071).

Sample size

The sample-size requirement was evaluated with reference to the primary hierarchical multiple regression analysis. The final regression model included four demographic covariate parameters and three core predictors. The required sample size was estimated using G*Power 3.1.9.7. An a priori power calculation for testing the incremental contribution of the three core predictors, assuming a small effect size of f² = 0.02, a two-sided significance level of 0.05, and statistical power of 0.80, indicated that at least 550 participants were required. The final analytic sample of 817 students therefore exceeded the minimum requirement for the primary regression analysis. The number of scale items and the need to exclude invalid questionnaires were also considered when determining the number of questionnaires distributed.

Measurement instruments

Social anxiety scale for adolescents

This study used the Chinese version of the Social Anxiety Scale for Adolescents (SAS-A), originally developed by La Greca and Lopez [32] and translated and validated for Chinese adolescents by Zhou et al. [43], to assess levels of SA among adolescents. The scale contains 22 items, of which 18 are formally scored and 4 are filler items. The scored items comprise three dimensions: Fear of Negative Evaluation (FNE; 8 items), Social Avoidance and Distress in New Situations (SAD-N; 6 items), and General Social Avoidance and Distress (SAD-G; 4 items). Items are rated on a 5-point Likert scale, ranging from 1 = “not at all true” to 5 = “very true.” The filler items are not included in the total score. Total scores range from 18 to 90, with higher scores indicating higher levels of SA. It should be noted that the SAS-A was used to assess SA symptom levels in adolescents and does not represent a clinical interview-based diagnosis of social anxiety disorder. In the present study, the Cronbach’s α coefficient for this scale was 0.882.

Physical activity rating scale-3

This study used the Physical Activity Rating Scale-3 (PARS-3), revised by Liang [44], to assess adolescents’ PE levels. The scale assesses exercise volume across three aspects: intensity, duration, and frequency. Exercise volume is calculated as intensity × duration × frequency. Intensity and frequency are each scored on a 5-point scale from 1 to 5, whereas duration is scored on a 5-point scale from 0 to 4. Total scores range from 0 to 100. In this study, the PARS-3 total score was used as the indicator of PE, with higher scores indicating higher levels of PE. The Cronbach’s α coefficient for this scale was 0.811.

Regulatory emotional self-efficacy scale

This study used the Chinese version of the Regulatory Emotional Self-Efficacy Scale (RESES), originally developed by Caprara et al. [19] and validated among Chinese junior middle school students by Huang et al. [45], to assess adolescents’ perceived confidence in their ability to regulate emotions. The scale consists of 12 items across three dimensions: managing despondency or distress (DES; 5 items), managing anger or irritation (ANG; 4 items), and expressing positive emotions (POS; 3 items). Items are rated on a 5-point Likert scale, ranging from 1 = “not well at all” to 5 = “very well.” In this study, the mean score across all items was used as the overall RESE indicator, with higher scores indicating stronger perceived confidence in one’s emotion-regulation ability. The Cronbach’s α coefficient for this scale was 0.843.

Social avoidance and distress scale

This study used the SAV subscale of the Chinese version of the Social Avoidance and Distress Scale (SADS), originally developed by Watson and Friend [31], and validated in Chinese student samples by Peng et al. [46], to assess adolescents’ avoidance tendencies in social situations. This dimension contains 14 items and uses a dichotomous “yes/no” response format, with both positively and reversely scored items. Items 2, 8, 13, 18, 21, 24, and 26 are positively scored, whereas items 4, 9, 17, 19, 22, 25, and 27 are reversely scored. For positively scored items, a “yes” response is scored as 1; for reversely scored items, a “no” response is scored as 1. All other responses are scored as 0. Total scores range from 0 to 14, with higher scores indicating higher levels of SAV. In the present study, the Cronbach’s α coefficient for this dimension was 0.938.

Statistical Analysis

Statistical analyses were performed using IBM SPSS Statistics 27.0. Before the main analyses, the data were examined for missing values, response distributions, and invalid response patterns. No missing values were present in the final analytic dataset. Descriptive statistics were calculated for demographic characteristics and the main study variables. Internal consistency was evaluated using Cronbach’s alpha. Common method bias was assessed using Harman’s single-factor test. The analysis included the 47 formally scored items from the four instruments; the four filler items of the Social Anxiety Scale for Adolescents were excluded. An unrotated principal component analysis was conducted, and the percentage of variance explained by the first factor was examined. The Kaiser–Meyer–Olkin statistic and Bartlett’s test of sphericity were also calculated. To describe the empirical factor structure in the present sample, principal component analysis with varimax rotation was conducted. Factors with eigenvalues greater than 1 were retained. A primary loading of at least 0.40 was considered acceptable, and potential cross-loading was defined as a secondary loading of at least 0.30 or a difference of less than 0.20 between the two largest loadings.

Zero-order Pearson correlations were used to describe the bivariate relationships among physical exercise, regulatory emotional self-efficacy, social avoidance, and social anxiety. The primary analysis was a two-step hierarchical multiple linear regression with social anxiety as the dependent variable. Sex, grade, and household registration were entered in Model 1 Sex was coded as 0 for boys and 1 for girls, and household registration was coded as 0 for urban and 1 for rural. Grade was represented by two dummy variables, with Grade 7 as the reference category. Physical exercise, regulatory emotional self-efficacy, and social avoidance were added in Model 2. Unstandardized coefficients, standardized coefficients, 95% confidence intervals, variance inflation factors, R², adjusted R², change in R², and the corresponding F-change statistic were reported.

The distributions of the main variables and the assumptions of the regression model were examined using descriptive indices, residual plots, normal probability plots, the Durbin–Watson statistic, variance inflation factors, and influence diagnostics, including studentized residuals, leverage values, and Cook’s distance. Because mild heteroscedasticity was detected, HC3 heteroscedasticity-consistent standard errors and confidence intervals were used for regression inference. Individual-level class identifiers were not collected, precluding class-level clustering adjustment, and the small number of participating schools limited the feasibility of a reliable school-level clustering sensitivity analysis.

Several supplementary analyses were conducted to evaluate the robustness and boundaries of the findings. First, because the physical exercise score was positively skewed, the main model was repeated using log(PE + 1), and the model was also re-estimated after excluding potentially influential observations. Second, to examine whether the association between social avoidance and total social anxiety was attributable solely to overlap in avoidance-related item content, the regression was repeated using fear of negative evaluation as an alternative outcome. Third, the three dimensions of regulatory emotional self-efficacy were entered simultaneously in a supplementary model. Fourth, sex-by-predictor interaction terms were tested after centering the continuous predictors. Finally, grade differences in social anxiety were treated as exploratory. Homogeneity of variance was assessed using Levene’s test, post hoc comparisons were conducted using Tukey’s procedure when the equal-variance assumption was met, and effect sizes were reported. All statistical tests were two-sided, with p < 0.05 considered statistically significant.

Results

Participant characteristics and data quality

A total of 817 students were included in the final analytic sample. The demographic characteristics of the participants are presented in Table 1. The sample included 412 boys (50.43%) and 405 girls (49.57%). Among the participants, 280 were in Grade 7 (34.27%), 260 were in Grade 8 (31.82%), and 277 were in Grade 9 (33.90%). In addition, 430 students had urban household registration (52.63%), and 387 had rural household registration (47.37%). No missing values were present in the final analytic dataset.

Table 1.

Demographic characteristics of the participants

Characteristic	Category	n	%
Sex	Boys	412	50.43
	Girls	405	49.57
Grade	Grade 7	280	34.27
	Grade 8	260	31.82
	Grade 9	277	33.90
Household registration	Urban	430	52.63
	Rural	387	47.37
Total		817	100.00

The Cronbach’s alpha coefficients for the principal scales and their dimensions ranged from 0.811 to 0.938, indicating satisfactory internal consistency. Harman’s single-factor test showed that the first unrotated component explained 24.214% of the total variance, suggesting that no single factor dominated the covariance among the formally scored items. The Kaiser–Meyer–Olkin value was 0.935. Bartlett’s test of sphericity was statistically significant, χ²(1081) = 19997.319, p < 0.001. Eight components with eigenvalues greater than 1 were retained and collectively explained 63.928% of the variance. After varimax rotation, the largest loading for each item ranged from 0.671 to 0.841. No item met the predefined criteria for cross-loading. The complete rotated component-loading matrix is provided in Supplementary Table S2.

Descriptive statistics and zero-order correlations

Descriptive statistics, internal consistency coefficients, and zero-order correlations are presented in Table 2. PE scores were positively skewed. The mean score was 23.25, the standard deviation was 26.48, and the median was 12.00. The interquartile range was 3.00 to 36.00, and scores ranged from 0 to 100. A total of 118 participants, representing 14.44% of the sample, had a PE score of 0. SA was negatively correlated with PE, r = − 0.320, p < 0.001, and with RESE, r = − 0.330, p < 0.001. In contrast, SA was positively correlated with SAV, r = 0.434, p < 0.001. PE was positively correlated with RESE and negatively correlated with SAV, whereas RESE was negatively correlated with SAV. All zero-order correlations were statistically significant at p < 0.001.

Table 2.

Descriptive statistics, internal consistency, and zero-order correlations

Variable	M ± SD	Skewness	Cronbach’s α	1	2	3	4
1. PE	23.25 ± 26.48	1.34	0.811	—
2. RESE	3.02 ± 0.74	−0.03	0.843	0.256***	—
3. SAV	6.30 ± 5.10	0.22	0.938	−0.303***	−0.315***	—
4. SA	53.15 ± 12.61	0.20	0.882	−0.320***	−0.330***	0.434***	—

Note. PE = physical exercise; RESE = regulatory emotional self-efficacy; SAV = social avoidance; SA = social anxiety. ***p < 0.001

Hierarchical regression analysis of SA

A two-step hierarchical multiple linear regression was conducted to examine the adjusted associations of PE, RESE, and SAV with SA. Sex, grade, and household registration were entered in Model 1 PE, RESE, and SAV were added in Model 2 The complete regression results are presented in Table 3. Model 1 explained 1.2% of the variance in SA. The model was statistically significant, F(4, 812) = 2.566, p = 0.037. After PE, RESE, and SAV were added in Model 2, the model explained an additional 25.3% of the variance in SA, ΔR² = 0.253. This increase was statistically significant, ΔF(3, 809) = 92.933, p < 0.001. The final model explained 26.6% of the variance in SA, R² = 0.266, adjusted R² = 0.259. In the final model, PE and RESE were negatively associated with SA, whereas SAV was positively associated with SA. All three associations remained statistically significant after adjustment for sex, grade, and household registration. All variance inflation factors were below 1.33, indicating no problematic multicollinearity. The standardized coefficients were not used to rank the theoretical importance of the three predictors because the measures differed in scale structure, reliability, and conceptual proximity to SA.

Table 3.

Hierarchical Multiple Regression Analysis of Social Anxiety Panel A. Model 1: Demographic Covariates. B. Model 2: Demographic Covariates and Core Predictors. Panel C. Model Summary

A
Variable	B	HC3 SE	β	95% CI	p	VIF
Intercept	51.931	0.949	—	[50.069, 53.793]	< 0.001	—
Female	−0.760	0.881	−0.030	[− 2.490, 0.969]	0.389	1.001
Grade 8	3.324	1.111	0.123	[1.145, 5.504]	0.003	1.317
Grade 9	1.637	1.042	0.061	[− 0.408, 3.682]	0.116	1.318
Rural household registration	−0.030	0.885	−0.001	[− 1.767, 1.707]	0.973	1.003
B
Variable	B	HC3 SE	β	95% CI	p	VIF
Intercept	58.271	2.020	—	[54.307, 62.236]	<0.001	—
Female	−0.320	0.762	−0.013	[−1.815, 1.176]	0.675	1.002
Grade 8	2.829	0.971	0.105	[0.924, 4.735]	0.004	1.318
Grade 9	1.760	0.900	0.066	[−0.006, 3.527]	0.051	1.321
Rural household registration	−0.312	0.768	−0.012	[−1.819, 1.194]	0.684	1.004
PE	−0.083	0.015	−0.174	[−0.113, −0.053]	<0.001	1.140
RESE	−3.104	0.538	−0.182	[−4.161, −2.048]	<0.001	1.151
SAV	0.795	0.081	0.322	[0.636, 0.954]	<0.001	1.181
C
Model statistic	Model 1	Model 2
R²	0.012	0.266
Adjusted R²	0.008	0.259
ΔR²	—	0.253
Model F	2.566	41.793
Model p	0.037	<0.001
ΔF	—	92.933
ΔF p	—	<0.001

Note. Boys,Grade 7,and urban household registration were the reference categories.B = unstandardized regression coefficient;HC3 SE = heteroscedasticity-consistent standard error;β=standardized regression coefficient;CI = confidence interval;VIF = variance inflation factor.The F-change statistic for Model 2 tested the incremental contribution of PE,RESE, and SAV

Robustness and supplementary analyses

The regression diagnostic results indicated that the residuals were independent, with a Durbin–Watson value of 2.075. The residual distribution did not show a substantial departure from normality because the Jarque–Bera test was not statistically significant, p = 0.478. The variance inflation factors ranged from 1.002 to 1.321. The Breusch–Pagan test indicated mild heteroscedasticity, p = 0.027. Therefore, HC3 heteroscedasticity-consistent standard errors and confidence intervals were used for regression inference. Two observations had absolute externally studentized residuals greater than 3, and six observations exceeded the predefined leverage threshold. Forty-five observations exceeded the screening threshold of 4/n for Cook’s distance. However, the maximum Cook’s distance was only 0.016, and no observation approached the conventional value of 1. A total of 49 observations met at least one of the predefined screening criteria.

The principal findings remained unchanged when the positively skewed PE score was replaced by log-transformed PE. PE and RESE remained negatively associated with SA, whereas SAV remained positively associated with SA. The same pattern was observed after the 49 potentially influential observations were excluded. To examine whether the association between SAV and total SA was attributable only to overlapping avoidance-related content, the regression was repeated using FNE as an alternative outcome. PE and RESE remained negatively associated with FNE, whereas SAV remained positively associated with FNE (β = 0.231, p < 0.001), compared with β = 0.322 for SAV in the total SA model.

The dimension-level analysis of RESE showed that efficacy in managing despondency or distress and efficacy in managing anger or irritation were negatively associated with SA. In contrast, the independent association between efficacy in expressing positive affect and SA was not statistically significant. Formal interaction tests did not provide evidence that sex modified the associations of PE, RESE, or SAV with SA. Adding the three interaction terms increased the explained variance by 0.3%, ΔR² = 0.003. This increase was not statistically significant, ΔF(3, 806) = 0.922, p = 0.430. None of the three individual interaction terms was statistically significant.

The principal robustness and supplementary results are summarized in Table 4. Complete results are provided in Supplementary Tables S3–S5.

Table 4.

Robustness and supplementary regression analyses

Analysis	Predictor	B	HC3 SE	β	95% CI	p
PE log-transformation	log(PE + 1)	−1.738	0.288	−0.195	[− 2.304, − 1.173]	< 0.001
	RESE	−3.023	0.537	−0.177	[− 4.077, − 1.970]	< 0.001
	SAV	0.777	0.080	0.314	[0.619, 0.935]	< 0.001
Exclusion of potentially influential observations, n = 768	PE	−0.091	0.014	−0.199	[− 0.118, − 0.064]	< 0.001
	RESE	−3.019	0.478	−0.188	[− 3.957, − 2.081]	< 0.001
	SAV	0.882	0.072	0.383	[0.740, 1.023]	< 0.001
Alternative outcome: FNE	PE	−0.029	0.010	−0.103	[− 0.049, − 0.010]	0.004
	RESE	−1.405	0.360	−0.138	[− 2.111, − 0.698]	< 0.001
	SAV	0.340	0.054	0.231	[0.235, 0.446]	< 0.001
RESE dimension-level model	DES	−1.439	0.383	−0.117	[− 2.191, − 0.686]	< 0.001
	ANG	−0.968	0.415	−0.078	[− 1.783, − 0.154]	0.020
	POS	−0.645	0.402	−0.052	[− 1.434, 0.144]	0.109
Sex interaction model	Sex × PE	0.019	0.031	—	[− 0.041, 0.079]	0.538
	Sex × RESE	−1.337	1.071	—	[− 3.440, 0.766]	0.212
	Sex × SAV	−0.191	0.162	—	[− 0.509, 0.128]	0.241

Note. All models controlled for sex, grade, and household registration. The continuous predictors were mean-centered before the interaction terms were calculated. FNE = fear of negative evaluation; DES = efficacy in managing despondency or distress; ANG = efficacy in managing anger or irritation; POS = efficacy in expressing positive affect

Exploratory grade differences in SA

Grade differences in SA were treated as exploratory because they were not part of the primary hypotheses. Levene’s test did not indicate unequal variances across the three grade groups, p = 0.244. The one-way analysis of variance showed a statistically significant grade difference, F(2, 814) = 4.766, p = 0.009. However, the magnitude of the difference was small, η² = 0.012, ω² = 0.009. Tukey-adjusted post hoc comparisons showed that Grade 8 students reported higher SA scores than Grade 7 students. The mean difference was 3.337 points, the adjusted p value was 0.006, and Cohen’s d was 0.261. The difference between Grade 7 and Grade 9 students was not statistically significant, with an adjusted p value of 0.273. The difference between Grade 8 and Grade 9 students was also not statistically significant, with an adjusted p value of 0.260. These findings indicate a small cross-sectional grade difference within the present sample. They should not be interpreted as evidence of age-related or pubertal developmental change. The complete grade-comparison results are provided in Supplementary Table S6.

Discussion

This study examined the hypothesized cross-sectional associations of PE, RESE, and SAV with SA among 817 junior high school students, together with exploratory analyses of sex-by-predictor interactions and grade differences in SA. The zero-order and adjusted results were consistent with H1–H3: higher PE and RESE were associated with lower SA, whereas higher SAV was associated with higher SA, and all three associations remained statistically significant after adjustment for sex, grade, and household registration. Formal interaction tests did not provide statistical evidence that sex modified the associations of PE, RESE, or SAV with SA. The directions of the principal associations were also unchanged when PE was log-transformed and when potentially influential observations were excluded, indicating that the main findings were not materially altered by the skewed PE distribution or by a limited number of observations. Overall, the findings characterize PE, RESE, and SAV as distinct concurrent correlates of SA rather than as components of a single causal or developmental pathway. The exploratory grade analysis identified a small difference in SA, with Grade 8 students reporting higher scores than Grade 7 students, whereas the other pairwise comparisons were not statistically significant. Because the analysis was cross-sectional and neither exact age nor pubertal status was assessed, this finding should be interpreted as a small, sample-specific grade difference rather than evidence of age-related or pubertal developmental change.

Consistent with H1, higher PE was associated with lower SA in both the zero-order and adjusted analyses. This finding is consistent with evidence from a large nationally representative sample of Finnish adolescents, in which social anxiety symptoms were associated with a lack of intense exercise, as well as with recent Chinese adolescent research reporting an inverse association between PE and SA [16, 47]. Recent longitudinal findings further indicate that the temporal ordering of the PE–SA association may be complex. In a two-wave study of 877 Chinese middle-school students using the PARS-3 and SAS-A, higher SA predicted lower subsequent physical activity, suggesting that adolescents with greater SA may also become less physically active over time [48]. Evidence from intervention research also suggests that the mental health relevance of exercise may depend partly on how exercise is delivered [49]. The negative association between PE and SA was also preserved when the positively skewed PE score was log-transformed and when potentially influential observations were excluded, suggesting that the direction of the finding was not materially dependent on the original PE distribution or on a limited number of observations. These findings suggest that PE may represent a meaningful behavioral correlate of SA among junior high school students. Although the cross-sectional design does not permit conclusions about temporal or causal direction, the consistency of the association across the primary and sensitivity analyses supports further longitudinal and intervention research. Future studies should examine whether the PE–SA association varies according to exercise intensity, modality, and participation context.

Consistent with H2, higher overall RESE was independently associated with lower SA after adjustment for PE, SAV, and demographic covariates. By focusing on perceived capability rather than the frequency of specific regulation strategies, RESE adds a belief-based perspective to the understanding of adolescent SA. This interpretation is supported by broader evidence linking stronger RESE with more favorable psychosocial adjustment and fewer internalizing difficulties [50–52]. Developmental evidence further suggests that perceived efficacy in managing negative emotions and expressing positive affect may have partly distinct foundations and should not be assumed to function identically [53]. The dimension-level analysis further showed that efficacy in managing despondency or distress and efficacy in managing anger or irritation were negatively associated with SA, whereas efficacy in expressing positive affect was not independently significant. This observed pattern is consistent with previous evidence that different components of RESE may show distinct associations with internalizing outcomes among Chinese young adolescents [54]. Together, the overall and dimension-level findings identify RESE as a meaningful psychological correlate of SA and show that examining its specific components can provide information beyond the overall score.

Consistent with H3, higher SAV was independently associated with higher SA after adjustment for PE, RESE, and demographic covariates. Because SAV reflects a tendency to withdraw from or avoid social contact, this finding adds a behavioral perspective to the understanding of adolescent SA beyond subjective fear and distress alone. Broader longitudinal research has identified persistent social-withdrawal trajectories characterized by higher anxiety and reduced social contact, supporting the developmental relevance of avoidance-related behavior during adolescence [55]. Related longitudinal evidence suggests that higher SA may be followed by poorer positive peer connections, indicating that anxiety and avoidance-related tendencies may have relevance for adolescents’ subsequent social functioning [56]. The supplementary analysis using FNE as an alternative outcome further clarified the SAV finding. SAV remained positively associated with FNE, with a standardized coefficient of 0.231 compared with 0.322 in the total SA model. Because FNE captures evaluative fear without avoidance-related items, this result suggests that the association of SAV with SA extended beyond shared avoidance content and was also related to adolescents’ concerns about negative evaluation. The relevance of this finding is supported by longitudinal evidence identifying FNE as a distinct prospective correlate of social anxiety symptoms during adolescence [57]. The smaller SAV coefficient in the FNE model is also consistent with the possibility that the conceptual and measurement proximity between SAV and total SA contributed to their stronger observed association. Together, these findings identify SAV as a meaningful behavioral correlate of SA and suggest that behavioral avoidance and evaluative fear provide related but non-identical information for characterizing adolescent social anxiety.

Beyond the three hypothesized associations, the exploratory analyses provided additional information on whether the observed associations varied by sex and whether SA differed across grades. Formal interaction tests did not indicate that sex modified the associations of PE, RESE, or SAV with SA. Within the present sample, this pattern suggests that the relevance of PE, RESE, and SAV to SA was not clearly confined to either boys or girls. Previous adolescent research has nevertheless documented sex differences in SA levels and developmental trajectories, indicating that mean-level sex differences and sex moderation of specific associations should be treated as related but distinct questions [58, 59]. The exploratory grade analysis showed that Grade 8 students reported higher SA than Grade 7 students, whereas Grade 9 students did not differ significantly from either group and the overall effect size was small. Accordingly, the observed pattern does not suggest a steady increase in SA across junior high school, but instead identifies a localized difference involving the middle grade in the present sample. Longitudinal research has similarly identified heterogeneous trajectories of SA symptoms across childhood and adolescence, suggesting that adolescents do not necessarily follow a single uniform developmental pattern [60]. Accordingly, the Grade 8 finding is better understood as a localized, sample-specific difference than as evidence of a general developmental peak in SA.

The present findings have practical relevance for school-based student support by identifying exercise behavior, emotion-regulation confidence, and social avoidance as three complementary domains associated with SA. Evidence from school-based cognitive-behavioral and emotion-regulation programs supports the feasibility of providing structured support for adolescent anxiety and emotional adjustment [61, 62]. Accordingly, support for students with stronger SAV may emphasize gradual engagement with avoided social situations, whereas support for those with lower RESE may focus on confidence in managing difficult negative emotions. PE may be considered a complementary health-promotion domain, although future intervention studies should determine whether its relevance to SA differs across exercise modalities, intensities, and settings. These domains may therefore inform differentiated school-based support while remaining complementary rather than interchangeable targets.

Limitations

This study has several limitations. First, the cross-sectional design does not permit conclusions regarding the temporal or causal directions of the associations among PE, RESE, SAV, and SA. Longitudinal and experimental studies are needed to examine how these variables change over time. Second, all variables were assessed using self-report measures and may therefore be affected by recall bias, social desirability, and common method variance. The PARS-3 assessed overall exercise volume but did not distinguish specific exercise modalities or participation contexts. In addition, SAV was conceptually close to the avoidance-related dimensions of the SAS-A; although the alternative FNE analysis suggested that shared avoidance content did not fully account for the association, this overlap may still have influenced its magnitude. Third, exact age and pubertal status were not assessed, limiting the interpretation of the observed grade difference. Finally, participants were recruited through convenience sampling from 16 classes in five schools within one city, which may limit the generalizability of the findings. Because class- and school-level clustering was not explicitly modeled, unaccounted within-class or within-school dependence may have affected the estimated standard errors and statistical inference. Future multisite studies should retain class-level identifiers and use analytic approaches that account for the nested data structure.

Conclusions

This cross-sectional study showed that lower PE, lower RESE, and higher SAV were independently associated with higher SA among junior high school students. Rather than representing a single developmental pathway, these findings indicate that exercise behavior, beliefs about emotion-regulation capability, and social avoidance provide complementary but non-interchangeable perspectives on adolescent SA. The supplementary analyses further provided a more nuanced view of RESE by revealing domain-specific patterns in its associations with SA, while also indicating that the association between SAV and SA cannot be attributed solely to shared avoidance-related content. Together, these findings may inform differentiated student support and provide a basis for longitudinal and intervention research examining how changes in these behavioral and psychological domains relate to subsequent changes in SA.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

PE

Physical exercise

SA

Social anxiety

RESE

Regulatory emotional self-efficacy

SAV

Social avoidance

M

Mean

SD

Standard deviation

Author contributions

DYH was responsible for questionnaire distribution, data analysis, and manuscript writing. LMY was responsible for the overall coordination of the study. ZKY and WWB revised the manuscript. All authors have read and approved the final version of the manuscript.

Funding

Not applicable.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

First, this study strictly adhered to the principles set out in the Declaration of Helsinki, and all methods were carried out in accordance with relevant guidelines and regulations. Second, this study was approved by the Ethics Committee of the Academic Committee of Capital University of Physical Education and Sports (2026A071). Finally, informed consent was obtained from all junior high school students and their guardians.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.