Metacognitive beliefs and academic performance in Chinese middle school students: A moderated mediation model of emotional dysregulation and anxiety

Abstract

Background

Academic performance during adolescence is shaped by the interplay of cognitive and emotional processes. According to the Self‐Regulatory Executive Function (S‐REF) model, maladaptive metacognitive beliefs (MB) contribute to anxiety, while deficits in emotion regulation may exacerbate this process. However, little is known about how specific MB dimensions interact with ED to influence anxiety and academic performance among Chinese adolescents.

Aims

This study investigated whether ED moderated the relationship between MB dimensions and anxiety and whether the indirect effect of MB on academic performance via anxiety was moderated by emotional dysregulation (ED) in Chinese adolescents.

Sample

A total of 353 middle school students (ages 12–15; 174 females) from Zhejiang Province, China, participated in the study.

Methods

Participants completed questionnaires on MB (MCQ‐30), ED (DERS‐16), and anxiety (DASS‐21 Anxiety Subscale). Academic performance was indexed using standardized examination scores. Latent moderated structural equation modelling (LMS) was used to test the moderated mediation effects.

Results

Results revealed that ED significantly moderated the positive relationships between negative beliefs, cognitive confidence and need for control with anxiety. Furthermore, an inconsistent mediation (suppression) effect was found, where the negative beliefs were positively associated with academic performance, while ED amplified the negative indirect effect of negative beliefs on academic performance via anxiety.

Conclusions

These findings underscore the importance of targeting both maladaptive MB and ED in school‐based interventions. Targeting specific MB components, such as negative beliefs, may be particularly beneficial for adolescents with poor emotion regulation capacities.

INTRODUCTION

In China, adolescents often experience heightened anxiety due to excessive academic workload, unrealistic expectations and intense competition (Sun et al., 2013; Zhao, 2015). Anxiety, characterized by apprehension, nervousness and somatic tension (Barlow, 2002), is rising globally in academic contexts (e.g., Azasu et al., 2024), with a particularly high prevalence reported among Chinese students (e.g., Yue et al., 2024). For example, 13.9% of Chinese middle school students meet the diagnostic criteria for anxiety, making it one of the most common mental health concerns in this population (Chen et al., 2023).

Although anxiety has been extensively studied in school contexts, existing literature has disproportionately focused on test anxiety. Within this broader landscape of school‐related anxiety, test anxiety represents a salient, situation‐specific form that is commonly divided into cognitive (worry) and physiological (emotionality) components (Cassady & Johnson, 2002; Liebert & Morris, 1967). Meta‐analyses have provided consistent evidence of a negative association between test anxiety and academic performance (Richardson et al., 2012; von der Embse et al., 2018), with the correlations for emotionality being consistently smaller than those for worry (von der Embse et al., 2018). Nevertheless, some recent work indicated that academic underperformance was driven not by worry itself but by related cognitive processes such as interference and low confidence (Schillinger et al., 2021), whereas a moderate level of emotionality may even facilitate performance (Chin et al., 2017).

While research on test anxiety and academic performance has yielded important insights, it may provide an incomplete account of students' emotional experiences in everyday school environments. In practice, students' anxiety in school extends well beyond examinations and academic testing. Middle school students are routinely exposed to multiple sources of stress. In mainland China, students' anxiety is shaped not only by academic demands but also by the broader developmental ecology in which they grow up, particularly parental academic expectations and competitive norms embedded in Confucian culture (Li et al., 2022; Stankov, 2010). Therefore, examining general, non‐trait anxiety allows for a more comprehensive understanding of students' emotion regulation in school settings and better captures the cumulative impact of daily academic and social stressors. Recent evidence suggested that general anxiety was negatively correlated with academic performance among Chinese middle school students, as heightened anxiety could potentially disrupt cognitive control, physiological discomfort and information processing. Accordingly, further identifying the underlying cognitive and emotional mechanisms is essential for informing early and targeted interventions (Wu et al., 2025).

The Self‐Regulatory Executive Function (S‐REF) model offers a comprehensive framework for understanding the cognitive‐affective processes that contribute to emotional distress, such as anxiety (Wells, 2009; Wells & Matthews, 1996). It suggests that psychological disorders arise from a cognitive–behavioural malfunction, the cognitive attentional syndrome (CAS), in which negative thoughts dominate attention and trigger maladaptive responses such as worry, rumination and threat monitoring. CAS results from the interaction of three interacting systems according to the S‐REF model: (1) low‐level, automatic processing units that generate intrusions; (2) a central S‐REF (control of action) level where appraisal, controlled process and strategic monitoring occur and (3) an upper level of metacognitive knowledge stored in long‐term memory (Wells, 2019; Wells & Matthews, 1996). While initially developed in clinical contexts, the model has increasingly been applied to non‐clinical samples to examine the cognitive roots of anxiety and impaired behavioural functioning, including academic outcomes (e.g., Huntley et al., 2023).

Under the S‐REF framework, the current study intended to explore how metacognitive beliefs (MB) and emotional dysregulation (ED) interactively relate to general, non‐trait anxiety and academic performance in Chinese adolescents. By identifying key cognitive‐emotional risk mechanisms, we also aimed to inform school‐based interventions to reduce anxiety and improve academic outcomes among middle school students in the future.

The indirect effect of MB on academic performance via anxiety

Metacognition refers to the evaluation, monitoring and regulation of one's own cognition, commonly described as “thinking about thinking” (Flavell, 1979). MB, or declarative knowledge about these processes, constitutes a core component of the self‐belief system in the S‐REF model (Wells, 1995; Wells, 2009). MB is typically classified into two broad categories that contain five interrelated yet distinct dimensions: positively valenced MB, which includes positive beliefs about worry and cognitive self‐consciousness, and negatively valenced MB, which includes negative beliefs about worry, cognitive confidence and need for control (Gerges et al., 2023; Wells, 2000; Wells & Cartwright‐Hatton, 2004). Positively valenced MB concerns the perceived utility of worry and meta‐awareness for regulation, whereas negatively valenced MB reflects pervasive self‐convictions about deficient control and the harmfulness of uncontrollability (Gerges et al., 2023).

According to the S‐REF model, MB is a major contributor to CAS, predisposing individuals to increased perseverative thinking, avoidance and suppression, thereby increasing their vulnerability to anxiety (Wells, 2009; Wells & Matthews, 1996). Negatively valenced MB, in particular, is central to emotional distress, such as anxiety, and forms the focus of empirical research within the S‐REF framework (Wells, 1995; Wells & Cartwright‐Hatton, 2004). Reflecting their clinical relevance, metacognitive therapy has been developed to target these beliefs to reduce CAS and strengthen emotion regulation ability (Wells, 2009), with evidence of its efficacy across conditions (Batebi et al., 2020; Normann et al., 2014; Philipp et al., 2019).

A robust empirical base links dysfunctional MB to elevated general anxiety in both clinical and community adolescent samples. A recent meta‐analysis confirmed this association across both clinical and non‐clinical samples of adolescents (Thingbak et al., 2024). Nordahl et al. (2019) found that MB explained 83% of the variance in generalized anxiety symptoms, and negatively valenced MB predicted anxiety in a Norwegian adult sample. Similarly, in a UK community sample, MB predicted general anxiety one month later (Capobianco et al., 2019).

Building on this foundation, researchers have extended these dynamics to the educational context. Studies using community adolescent samples have suggested that anxiety may mediate the negative relationship between MB and academic performance. In a network analysis of French adolescents, Cécillon et al. (2024) observed that MB was negatively correlated with trait anxiety and academic performance, and anxiety was negatively linked to academic performance. Among British undergraduate students, Huntley et al. (2023) found that negative beliefs positively predicted higher test anxiety, whereas anxiety was negatively associated with exam scores. However, previous research tended to focus on the global MB score (e.g., Capobianco et al., 2019; Cécillon et al., 2024) and failed to examine how specific MB dimensions differentially impact academic performance through anxiety. Moreover, the literature has disproportionately focused on test anxiety in Western samples, which does not capture the broader, context‐general anxiety among Chinese adolescents. To address these gaps, the current study investigated the indirect effects of each MB dimension on academic performance via general, non‐trait anxiety among Chinese middle school students and explored the roles of different MB dimensions in the pathways.

The moderating role of ED in the relationship between MB and anxiety

Adolescents are especially vulnerable to ED due to rapid socio‐emotional development and frequent use of maladaptive strategies (Lougheed & Hollenstein, 2012; Soto et al., 2011). Emotion regulation refers to the processes by which individuals monitor, evaluate and modify emotional responses to meet their goals, whereas ED reflects persistent difficulties in these processes (Gross & Jazaieri, 2014; Gross & Thompson, 2007). In line with recent conceptualizations, this study adopts a broad definition of ED that includes the problematic intensity, duration and frequency of emotions, maladaptive strategies and general regulatory difficulties (Gross & Jazaieri, 2014; Schneider et al., 2016). ED is a key risk factor for various psychopathologies among adolescents, particularly anxiety (Aldao et al., 2010; Bender et al., 2012; Iannattone et al., 2024). Longitudinal evidence from Chinese student samples indicated that ED predicts elevated anxiety levels (Chan et al., 2023), whereas adaptive regulation strategies reduce anxiety (Zou et al., 2025).

The S‐REF model identifies the control of the action system as the site of conscious, voluntary and strategic processing, where individuals perform appraisals of mental intrusions and initiate action‐control strategies to resolve perceived discrepancies between the current and desired emotional states (Wells, 2009; Wells & Matthews, 1996). Failures in this system directly contribute to the onset and maintenance of anxiety (Wells, 2009). Consistent with this framework, emotion regulation is an aspect of the executive self‐control system as it utilizes the top‐down, volitional capacities of executive function to monitor internal affect and implement cognitive strategies aimed at resolving discrepancies between an individual's current and desired emotional states (Blair & Ursache, 2011; Mattioni et al., 2025). Therefore, ED represents the dysfunctional action of emotional control. Importantly, the S‐REF model posits an interdependent relationship between the self‐control system and MB: effective self‐regulation buffers the negative outcomes of dysfunctional MB, whereas dysregulation processes (e.g., worry and suppression) reinforce MB and promote anxiety (Wells, 2009; Wells & Matthews, 1996). This interaction is particularly relevant in the context of emotion regulation, as Ford and Gross (2019) highlighted, by studying the potential interplay between beliefs about the uncontrollability and ineffectiveness of emotions and emotion regulation.

Although few studies have directly tested the interactive roles of MB and ED, indirect evidence supports their potential interaction. Ochsner et al. (2004) supported the interplay between cognitive control and emotional regulation in determining the effectiveness of emotion control, as high‐level cognitive processes play a crucial role in modulating emotion regulation. Hien et al. (2017) showed that individuals with high ED levels benefited less from relapse prevention therapy, a cognitive–behavioural intervention that relies on monitoring and restructuring maladaptive thought patterns. Hu et al. (2025) found that the interaction between emotional growth mindsets (viewing emotion as malleable) and perceived stress controllability weakened the positive relationship between perceived stress intensity and anxiety symptoms. Notably, the mindset that emotions are immutable is a core feature of ED, while perceived stress controllability is associated with MB, particularly the negative belief dimension. This suggests that ED may also heighten the MB–anxiety association by impairing cognitive control and promoting perseverative thinking.

Although direct tests of the interaction between MB and ED are limited, several studies have explored them as independent predictors of anxiety (e.g., Salguero et al., 2019). While Gerges et al. (2023) reported that ED amplified the link between MB and orthorexia nervosa, no such interactive effect emerged for trait anxiety (Cécillon et al., 2024). These mixed findings underscore the need for further investigation into the interaction between ED and MB, particularly in the context of anxiety.

While few studies have specifically examined the moderating role of ED in the internal cognitive pathways (e.g., MB) leading to anxiety, previous literature has widely explained how ED magnifies the effects of external stressors. Individuals with high ED experience stronger effects of adverse life events on anxiety, as they are more likely to become overwhelmed and engage in perseverative thinking under stress (Boyes et al., 2016; Durham et al., 2025; Graham et al., 2015). This dysfunctional top–down control mirrors CAS described in the S‐REF model (Wells & Matthews, 1996), suggesting that similar mechanisms may operate when internal cognitive risks, such as maladaptive MB, give rise to anxiety. Furthermore, Besson et al. (2023) found that ED impairs the ability to direct attention and appraise emotional cues, representing a deficit that mirrors the attentional rigidity underlying CAS. Together, these findings imply that ED may moderate the pathways through which maladaptive cognition contributes to anxiety.

Given the mixed findings, the limited attention to general, non‐trait anxiety, which is more sensitive to CAS activation, and the distinctive high‐stakes educational context faced by Chinese adolescents, the present study tests whether ED moderates the association between maladaptive MB and anxiety in this population. This moderation test clarifies when executive regulatory difficulties amplify the link between MB and anxiety, thereby specifying a core boundary condition of the S‐REF model in Chinese school settings.

The moderating role of ED on the indirect effect of MB on academic performance via anxiety

Emotion regulation has been widely examined in the educational context and has been indicated to influence children's academic achievement (e.g., Chen‐Bouck et al., 2024; Djambazova‐Popordanoska, 2016). ED has been consistently linked to impaired cognitive functioning and lower academic achievement (Soltani Nezhad & Delroba, 2024; Usán Supervía & Quílez Robres, 2021). Proposed mechanisms include diminished academic self‐efficacy (Usán Supervía & Quílez Robres, 2021), executive dysfunction (Cécillon et al., 2024) and maladaptive externalizing behaviour (Graziano et al., 2007).

Building on this literature and the S‐REF framework, we consider a moderated‐mediation process linking ED, MB, anxiety and academic performance. Specifically, if MB triggers CAS and subsequent anxiety, this state of distress will deplete the limited‐capacity attentional resources and thereby undermine academic performance. Crucially, the S‐REF model highlights an interactive effect between MB and ED as a failure in self‐control: we hypothesize that higher ED may amplify the translation of maladaptive MB into anxiety and its downstream academic costs by making individuals less able to inhibit or disengage from CAS. Conversely, adolescents with low ED may utilize more flexible attentional control to mitigate the MB–anxiety association, thereby shielding their academic functioning from the downstream effects of anxiety.

Accordingly, the present study tests whether ED moderates the indirect effect of MB on academic performance via anxiety, refining S‐REF by specifying when metacognitive risk turns into academic underachievement. Clarifying this interaction is crucial for understanding the cognitive–emotional mechanisms that influence learning outcomes and for informing interventions aimed at reducing anxiety and enhancing academic success among Chinese adolescents.

The present study

Guided by the S‐REF framework, the current study aims to present an integrated model to examine the interplay between ED, MB, anxiety and academic performance. The model is specified on theoretical grounds (Wells & Matthews, 1996) and is informed by prior findings (e.g., Chan et al., 2023; Nordahl et al., 2023). However, given the cross‐sectional design, the analyses address associations rather than temporal or causal effects. This study seeks to explore how MB and ED influence anxiety and, in turn, impair academic performance in adolescents. We also intend to test each MB dimension separately. Our hypotheses are as follows:

Hypothesis 1

Anxiety, ED and all dimensions of MB are positively correlated with each other.

Hypothesis 2

ED moderates the association between MB and anxiety, such that higher levels of ED strengthen the positive relationship between MB and anxiety.

Hypothesis 3

ED moderates the indirect effect of MB on academic performance via anxiety, such that this indirect effect becomes more negative as ED increases.

For Hypothesis 2 and 3, we do not specify dimension‐level predictions for MB. This decision is due to the limited prior evidence on whether and which specific MB dimensions are most strongly implicated in the ED–anxiety pathway and its downstream academic consequences. Accordingly, analyses involving MB dimensions are treated as exploratory.

MATERIALS AND METHODS

Participants and procedure

Ethical approval for this study was obtained from the Research Ethics Committee of the authors' institution before data collection (Ethics Number: EF20241128001). Participants were recruited using convenience sampling from a public middle school in Zhejiang Province, China. Data were collected in mid‐to‐late December 2024 via a voluntary self‐report survey. In China, formal education typically follows a 6–3‐3 structure: children attend primary school for 6 years (approximately ages 6–12), middle school for 3 years (ages 12–15) and high school for 3 years (ages 15–18). Before data collection, the project was introduced to the students and their parents and guardians by the psychology teachers and class headteachers, following approval from the school administrator. Students and their parents or other guardians were fully informed of the study's purpose and procedures, their right to withdraw at any time without consequences, and that participation was voluntary, anonymous, and strictly confidential. After informed consent forms were obtained from both students and their parents and guardians, the students completed an online questionnaire via Credamo during regular school hours, which took an average of 20 min. To ensure practical implementation and maintain consistency across participants, the questionnaires were administered in a fixed order.

No missing data were observed. We followed the data screening methods suggested by DeSimone et al. (2015). Two attention check items (e.g., “Please choose the item ‘agree’”) were used in the online survey as direct screening tools to exclude respondents who failed to follow instructions. In addition, we used the archival data screening method for response time to exclude subjects with a response time of less than 2 s per item (Huang et al., 2012). We did not exclude participants based on demographic or performance‐related criteria. After excluding invalid responses, 353 valid responses (174 females; overall response rate of 76%) were obtained, with ages ranging from 12 to 15 years (M _age = 12.79 years, SD = .70). Power analysis was conducted using Soper's (2025) A priori Sample Size Calculator for Structural Equation Model (SEM) showed that we needed 184 participants to achieve 80% power to detect a medium effect size (δ = .30) with α = .05 and model complexity consisting of nine latent variables and 32 observed variables (Cohen, 1977; Soper, 2025). Our sample (N = 353) exceeded this threshold.

There was a significant difference between the excluded and included participants in sex distribution, χ²(1, N = 464) = 11.84, p < .001, with a higher proportion of males in the excluded group (69.4%) compared to the included group (50.1%). However, there were no significant differences in age between the excluded (M = 12.70, SD = .73) and included participants (M = 12.79, SD = .70), t(176.66) = 1.11, p = .268, 95% CI [−.07, .24]. Excluded participants had significantly lower academic performance (M = 328.38, SD = 87.97) than included participants (M = 360.87, SD = 66.24), t(151.23) = 3.58, p < .001, 95% CI [14.58, 50.40].

Measures

Anxiety subscale of the depression anxiety and stress Scales‐21 (DASS‐21)

General, non‐trait anxiety was assessed using the Chinese version of the DASS‐21 Anxiety subscale (Gong et al., 2010), adapted from Lovibond and Lovibond (1995). The DASS‐21 comprises Depression, Anxiety, and Stress factors and has demonstrated measurement invariance across gender and grade among Chinese adolescents (Luo et al., 2025). In addition, previous research supports interpreting the DASS‐21 subscales independently when the research question targets a specific symptom domain (Henry & Crawford, 2005). Accordingly, we used the Anxiety subscale as a stand‐alone measure (e.g., “I felt I was close to panic”). Prior studies have reported satisfactory structural validity and reliability for the DASS‐21 Anxiety subscale in Chinese adolescent samples (Luo et al., 2025; Mellor et al., 2015). Participants rated their experiences over the past week on seven items using a 4‐point scale ranging from 1 (“least likely”) to 4 (“most likely”). Confirmatory factor analysis (CFA) showed that the one‐factor model of the DASS‐21 Anxiety subscale showed excellent fit indices (χ²(14) = 13.95, p = .454, CFI = 1.00, TLI = 1.00, RMSEA = .00, 90% CI [.00, .05], SRMR = .02). The anxiety subscale showed satisfactory internal consistency in the current sample (Cronbach's α = .83).

The metacognitions Questionnaire‐30 (MCQ‐30)

MB was measured using the Chinese version of the Metacognitions Questionnaire‐30 (MCQ‐30; Zhang et al., 2020), adapted from the original scale developed by Wells and Cartwright‐Hatton (2004). This instrument was chosen because it provides an efficient yet comprehensive assessment of the five key dimensions of MB (Wells & Cartwright‐Hatton, 2004). The MCQ‐30 consists of 30 items across five subscales: (1) positive beliefs about worry (e.g., “Worrying helps me to avoid problems in the future”); (2) negative beliefs about worry (e.g., “My worrying is dangerous for me”); (3) (lack of) cognitive confidence (e.g., “My memory can mislead me at times”); (4) beliefs about the need to control thoughts (e.g., “If I could not control my thoughts, I would not be able to function”) and (5) cognitive self‐consciousness (e.g., “I constantly examine my thoughts”). Participants rated each item on a 5‐point Likert scale ranging from 1 (“totally disagree”) to 5 (“strongly agree”). Higher scores indicate greater dysfunctional MB. The Chinese version of the MCQ‐30 has demonstrated satisfactory psychometric properties among Chinese students (Zhang et al., 2020).

CFA of each dimension of the Chinese version of the MCQ‐30 showed that the need for control, positive beliefs and cognitive self‐consciousness subscales had excellent model fits without modification. The negative belief subscale showed an excellent model fit after modification. Consequently, negative beliefs showed a good fit to the data, χ ² (8) = 11.58, p = .171, CFI = .98, TLI = .97, RMSEA = .04, 90% CI [.00, .08], SRMR = .03. Positive beliefs showed an acceptable fit to the data, χ ² (9) = 22.29, p = .008, CFI = .96, TLI = .93, RMSEA = .07, 90% CI [.03, .10], SRMR = .04. Need for control showed an acceptable fit to the data, χ ² (9) = 21.30, p = .011, CFI = .96, TLI = .93, RMSEA = .06, 90% CI [.03, .10], SRMR = .04. The cognitive confidence dimension showed an excellent fit to the data, χ ² (8) = 15.89, p = .044, CFI = .98, TLI = .97, RMSEA = .05, 90% CI [.01, .09], SRMR = .03. The cognitive self‐consciousness dimension showed an excellent fit to the data, χ ² (9) = 7.53, p = .582, CFI = 1.00, TLI = 1.02, RMSEA = .00, 90% CI [.00, .05], SRMR = .02. The Chinese version of the MCQ‐30 showed satisfactory internal consistency in the current sample regarding all MB dimensions: positive beliefs (Cronbach's α = .83), negative beliefs (Cronbach's α = .85), need for control (Cronbach's α = .73), cognitive confidence (Cronbach's α = .82), cognitive self‐consciousness (Cronbach's α = .75).

Difficulties in emotion regulation scale – 16 item version (DERS‐16)

ED was measured using the Chinese version of the 16‐item Difficulties in Emotion Regulation Scale (DERS‐16; Xiong et al., 2025), which was adapted from the original scale developed by Bjureberg et al. (2016). We used the DERS‐16 because it is a widely used and time‐efficient measure that demonstrates psychometric properties comparable to those of the original scale (Bjureberg et al., 2016; Demirpence Secinti & Sen, 2023). Participants rated each item (e.g., “When I'm upset, I become out of control”) on a 5‐point Likert scale ranging from 1 (“almost never”) to 5 (“almost always”), with higher scores indicating higher ED. The Chinese version of the DERS‐16 has demonstrated robust psychometric properties in Chinese adolescents (Xiong et al., 2025). The five‐factor model showed excellent fit indices in the current study (χ ² (99) = 250.40, p < .001, CFI = .94, TLI = .92, RMSEA = .07, 90% CI [.06, .08], SRMR = .05). The Chinese version of the DERS‐16 demonstrated satisfactory internal consistency in the current study (Cronbach's α = .94).

Academic performance

Academic performance was assessed using students' final standardized exam scores from the first semester of the 2024–2025 academic year, a widely recognized and robust indicator of academic achievement (e.g., Herrera et al., 2020; Zhang et al., 2025). The standardized exam evaluated performance across five core subjects: Chinese, Mathematics, English, Science and Sociology. Individual scores were calculated by summing the results across these subjects and then standardized as z‐scores for analysis. The exams were administered within one month of completing the online survey.

Measure of demographic information

Participants provided self‐reported demographic information, including sex (male or female), age, grade level, ethnicity, socio‐economic status and only‐child status. Socioeconomic status was assessed via parental education, a widely accepted indicator of SES in psychological and educational research (Bradley & Corwyn, 2002; Sirin, 2005). Participants separately reported the highest level of formal education attained by their mother and father using a 4‐point Likert scale. To create a composite SES indicator, the scores for mothers' and fathers' education were averaged, with higher scores indicating higher SES levels. Meanwhile, the current study collected participants' sex, defined as a set of biological attributes associated with physical and physiological characteristics, such as chromosomes, hormone function and reproductive anatomy (Heidari et al., 2016). The current study collected participants' sex rather than gender identity, primarily due to cultural and contextual considerations within the Chinese school system.

Statistical analysis

We first computed skewness and kurtosis for the data and then computed descriptive statistics (means, standard deviations, correlations) for all variables using IBM SPSS 26 (IBM Corp., 2019). CFA was conducted with the robust maximum likelihood estimator (MLR) in Mplus 8.3 (Muthén & Muthén, 1998–2017) to validate the measurement structure of each construct.

The model fits of the measurement models were evaluated based on established criteria (Hu & Bentler, 1998), including the chi‐square statistic, comparative fit index (CFI), Tucker–Lewis index (TLI), root mean square error of approximation (RMSEA) and standardized root mean square residual (SRMR). A CFI and TLI value of ≥.90 was considered a good fit; RMSEA ≤.06 (Bentler, 1990; Hu & Bentler, 1998) and SRMR ≤.10 (Kline, 2011; MacCallum et al., 1996) were considered indicative of an acceptable model fit.

To examine whether ED moderated (a) the relationship between each MB dimension and anxiety, and (b) the indirect effect of MB on academic performance via anxiety, we specified five latent moderated mediation models with the latent moderated structural equations (LMS) method (Klein & Moosbrugger, 2000; Stride et al., 2015). Since LMS does not provide conventional fit indices, we used a two‐step approach: first, we established well‐fitting baseline models (Model 0) with measurement and structural paths but no interactions; then, we estimated five moderated mediation models (Model 1) including latent interactions (Maslowsky et al., 2015; Stride et al., 2015). The fit of Model 0 and Model 1 was compared using a log‐likelihood ratio test, following the LMS approach, to determine whether the interaction significantly improved the model fit (Maslowsky et al., 2015).

Model 0 without the interaction term between ED and MB dimension showed a good fit for the models of negative beliefs, χ ² (103) = 819.54, p < .001, CFI = .92, TLI = .91, RMSEA = .05, 90% CI [.04, .05], SRMR = .09, cognitive confidence, χ ² (103) = 824.11, p < .001, CFI = .92, TLI = .91, RMSEA = .05, 90% CI [.04, .05], SRMR = .09, need for control, χ ² (102) = 831.87, p < .001, CFI = .91, TLI = .90, RMSEA = .05, 90% CI [.04, .05], SRMR = .09, positive beliefs, χ ² (104) = 754.08, p < .001, CFI = .93, TLI = .93, RMSEA = .04, 90% CI = [.04, .05], SRMR = .09, cognitive self‐consciousness, χ ² (101) = 791.59, p < .001, CFI = .92, TLI = .91, RMSEA = .05, 90% CI [.04, .05], SRMR = .09.

The log‐likelihood ratio test revealed that Model 1 significantly improved the model fit compared to Model 0 when negative beliefs (χ ² (1) = 41.80, p < .001), cognitive confidence (χ ² (1) = 60.02, p < .001), need for control (χ ² (1) = 43.64, p < .001), positive beliefs (χ ² (1) = 9.84, p < .001) and cognitive self‐consciousness (χ ² (1) = 14.22, p < .001) were treated as independent variables.

Simple slopes were plotted to illustrate the relationship between each MB dimension and anxiety at low (−1 SD), mean and high (+1 SD) levels of ED (Aiken & West, 1991).

RESULTS

Descriptive statistics and correlation analysis

The descriptive statistics and correlations for all variables are presented in Table 1. Note that zero‐order correlations are based on observed scores and may be attenuated by measurement errors. Subsequent SEM analyses estimated latent constructs and conditional relations while accounting for covariates and shared variance among predictors. All variables met the recommended cutoffs of ±2 for skewness and ± 7 for kurtosis (Hair et al., 2010; Kim, 2013; Kline, 2011). Therefore, the data distribution did not significantly deviate from normality.

TABLE 1.

Descriptive statistics and bivariate correlations among study variables (N = 353).

Variables	M	SD	Skewness	Kurtosis	1	2	3	4	5	6	7	8	9	10
1. Sex	–	–	–	–	–
2. Age	12.79	.70	.36	−.71	.01	–
3. Negative Beliefs	16.16	5.72	.19	−.65	.22***	.04	–
4. Positive Beliefs	15.04	4.90	.04	−.42	.05	.11*	.40***	–
5. Cognitive Confidence	15.35	5.23	.34	−.16	.22***	.08	.58***	.40***	–
6. Cognitive Self‐Consciousness	21.69	4.34	−.29	−.07	−.05	.03	.34***	.40***	.19***	–
7. Need for Control	14.83	4.66	.45	.12	.02	.11*	.55***	.32***	.57***	.28***	–
8. Emotional Dysregulation	35.86	14.83	.68	−.22	.12*	.07	.67***	.25***	.54***	.22***	.57***	–
9. Anxiety	10.86	4.18	1.52	2.36	.15**	.04	.59***	.26***	.56***	.21***	.47***	.72***	–
10. Academic Performance	360.87	66.24	−.73	.03	.07	.02	.06	.16**	−.03	.19***	.05	−.02	−.09	–

Note: Values represent means (M), standard deviations (SD), skewness and kurtosis for continuous variables. Sex was coded as 0 = Male and 1 = Female; correlations involving Sex are point‐biserial. Possible score ranges: Negative Beliefs = 6–30, Positive Beliefs = 6–30, Cognitive Confidence = 6–30, Cognitive Self‐Consciousness = 6–30, Need for Control = 6–30, Emotional Dysregulation = 16–80, Anxiety = 7–28, Academic Performance = 0–500.

^* p < .05;

^** p < .01;

^*** p < .001.

The correlation matrix (Table 1) shows that all dimensions of MB are intercorrelated (ps < .001). Each MB dimension was also significantly positively associated with ED and anxiety (ps < .001), with correlation coefficients ranging from r = .22 to .67 for ED and from r = .21 to .59 for anxiety. Among all dimensions of MB, negative beliefs showed relatively large associations with both ED (r = .67, p < .001) and anxiety (r = .59, p < .001). Additionally, there was a significant positive correlation between ED and anxiety (r = .72, p < .001).

Significant correlations were found between academic performance and specific MB dimensions, namely positive beliefs (r = .16, p = .002) and cognitive self‐consciousness (r = .19, p < .001), but not with the other dimensions (ps > .05). Meanwhile, academic performance was not significantly correlated with ED (r = −.02, p = .703) or with anxiety (r = −.09, p = .106).

In addition, point‐biserial correlations revealed significant sex differences in several variables. Girls scored higher than boys on negative beliefs, r = .22, p < .001; cognitive confidence, r = .22, p < .001; ED, r = .12, p = .023 and anxiety, r = .15, p = .004. Age was positively correlated with positive beliefs (r = .11, p = .040) and need for control (r = .11, p = .050).

Moderation analysis

The moderation results are presented in Table 2. Table 3 presents the simple slope results. For dimensions showing a significant MB × ED interaction, we graphed simple slopes at low (−1 SD), mean and high (+1 SD) ED levels (Figure 1).

TABLE 2.

Latent moderated structural equation model predicting anxiety and academic performance from MB, ED and their interaction.

MB dimension	Negative beliefs				Positive beliefs				Cognitive confidence				Cognitive self‐consciousness				Need for control
Parameter	b	SE	t	95% CI	b	SE	t	95% CI	b	SE	t	95% CI	b	SE	t	95% CI	b	SE	t	95% CI
Anxiety (DV)
MB dimension	.20	.07	2.87**	[.06, .33]	.05	.06	.82	[−.06, .16]	.18	.05	3.45**	[.08, .29]	.10	.08	1.27	[−.06, .26]	.09	.06	1.53	[−.03, .20]
ED	.25	.06	4.44***	[.14, .36]	.42	.05	7.89***	[.31, .52]	.27	.06	4.49***	[.15, .38]	.39	.06	6.32***	[.27, .51]	.32	.07	4.80***	[.19, .45]
MB dimension × ED	.21	.05	3.94***	[.11, .32]	.12	.08	1.46	[−.04, .29]	.23	.05	4.61***	[.13, .32]	.21	.11	1.84	[−.01, .42]	.21	.05	4.58***	[.12, .30]
Academic Performance (DV)
Anxiety	−.49	.21	−2.36*	[−.89, −.08]	−.31	.13	−2.35*	[−.57, −.05]	−.28	.20	−1.40	[−.67, .11]	−.37	.14	−2.68**	[−.64, −.10]	−.35	.18	−1.92	[−.71, .01]
MB dimension	.28	.14	2.06*	[.01, .55]	.34	.09	3.58***	[.15, .52]	.07	.11	.65	[−.15, .29]	.40	.10	3.90***	[.20, .61]	.18	.12	1.54	[−.05, .41]
Age	.03	.07	.37	[−.11, .16]	.00	.07	.01	[−.13, .13]	.02	.07	.34	[−.11, .16]	.02	.07	.32	[−.11, .15]	.02	.07	.26	[−.12, .15]
Sex	.12	.10	1.24	[−.07, .32]	.15	.10	1.58	[−.04, .34]	.15	.10	1.48	[−.05, .34]	.18	.10	1.91	[−.01, .37]	.16	.10	1.62	[−.03, .35]

Note: Values represent unstandardized regression coefficients (b), standard errors (SE), t values and 95% confidence intervals (CIs) from latent moderated structural equation models. Sex was coded as 0 = Male and 1 = Female.

Abbreviations: DV, dependent variable; ED, emotional dysregulation; MB, metacognitive beliefs.

^* p < .05;

^** p < .01;

^*** p < .001.

TABLE 3.

Simple slopes analysis for the moderating effect of ED on the relationship between MB and anxiety.

Metacognitive belief dimension	Low ED (−1 SD)				Mean ED				High ED (+1 SD)
	b	SE	t	95% CI	b	SE	t	95% CI	b	SE	t	95% CI
Negative Beliefs	−.01	.06	−.24	[−.12, .09]	.20	.07	2.87**	[.06, .33]	.41	.11	3.70***	[.19, .63]
Positive Beliefs	−.08	.05	−1.72	[−.17, .01]	.05	.06	.82	[−.06, .16]	.17	.14	1.24	[−.10, .44]
Cognitive Confidence	−.04	.04	−1.08	[−.12, .03]	.18	.05	3.45**	[.08, .29]	.41	.10	4.33***	[.22, .60]
Cognitive Self‐Consciousness	−.10	.05	−2.25*	[−.19, −.01]	.10	.08	1.27	[−.06, .26]	.31	.19	1.62	[−.07, .68]
Need for Control	−.12	.05	−2.36*	[−.22, −.02]	.09	.06	1.53	[−.03, .20]	.30	.09	3.27**	[.12, .48]

Note: Unstandardized regression coefficients (b) are reported. Simple slopes were calculated at ED = low (−1 SD), mean, and high (+1 SD).

Abbreviations: ED, emotional dysregulation; MB, metacognitive beliefs; SE, standard error.

^* p < .05;

^** p < .01;

^*** p < .001.

FIGURE 1.

Simple slopes for MB dimensions interacting with ED in relation to anxiety. ED, emotional dysregulation; MB, Metacognitive Beliefs. Panels display (a) Negative Beliefs × ED, (b) Cognitive Confidence × ED and (c) Need for Control × ED. Lines show estimated anxiety at ED = low (−1 SD), mean and high (+1 SD).

Negative beliefs

The latent interaction between ED and negative beliefs in relation to anxiety was significant (b = .21, SE = .05, t = 3.94, p < .001, 95% CI [.11, .32]). Specifically, higher levels of ED heightened the relationship between negative beliefs and anxiety. As shown in Figure 1, simple slopes analysis revealed no significant relationship at the low ED level (b = −.01, SE = .06, t = −.24, p = .812, 95% CI [−.12, .09]), a significant positive association at the mean ED level (b = .20, SE = .07, t = 2.87, p = .004, 95% CI [.06, .33]) and the high ED level (b = .41, SE = .11, t = 3.70, p < .001, 95% CI [.19, .63]).

Cognitive confidence

A significant moderating effect of ED on the relationship between cognitive confidence and anxiety was observed (b = .23, SE = .05, t = 4.61, p < .001, 95% CI [.13, .32]). Simple slopes (Table 3; Figure 1b) showed that the association was not significant at low ED (b = −.04, SE = .04, t = −1.08, p = .289, 95% CI [−.12, .03]), but was significant at mean ED (b = .18, SE = .05, t = 3.45, p = .001, 95% CI [.08, .29]) and high ED (b = .41, SE = .10, t = 4.33, p < .001, 95% CI [.22, .60]).

Need for control

ED significantly moderated the relationship between need for control and anxiety (b = .21, SE = .05, t = 4.58, p < .001, 95% CI [.12, .30]). Simple slope analyses (Table 3; Figure 1c) indicated a crossover pattern: at low ED, need for control was significantly negatively associated with anxiety (b = −.12, SE = .05, t = −2.36, p < .05, 95% CI [−.22, −.02]); at mean ED, the association was not significant (b = .09, SE = .06, t = 1.53, p = .13, 95% CI [−.03, .20]) and at high ED, the association was significantly positive (b = .30, SE = .09, t = 3.27, p = .001, 95% CI [.12, .48]).

Positive beliefs

In contrast, there was no significant moderating effect of ED on the relationship between positive beliefs and anxiety (b = .12, SE = .08, t = 1.46, p = .146, 95% CI [−.04, .29]). Simple slope analyses (Table 3) confirmed that the relationship between these MB dimensions and anxiety was not significant at any ED level (ps > .05).

Cognitive self‐consciousness

There was no significant moderating effect of ED on the relationship between cognitive self‐consciousness and anxiety (b = .21, SE = .11, t = 1.84, p = .064, 95% CI [−.01, .42]). Simple slope analyses (Table 3) showed a significant negative association at low ED (b = −.10, SE = .05, t = −2.25, p = .03, 95% CI [−.19, −.01]), whereas slopes at mean and high ED were not significant (all p > .05).

Moderated mediation analysis

Moderated mediation analyses were conducted to examine how ED influenced the indirect effect of each MB dimension on academic performance through anxiety. The results of the moderated mediation model are presented in Table 2, and the conditional effects are presented in Table 4.

TABLE 4.

Direct effect, conditional indirect and total effects of MB on academic performance through anxiety at different levels of ED.

MB dimension	Effect type	ED level	b	SE	t	95% CI
Negative Beliefs	IMM	–	−.10	.05	−2.03*	[−.20, −.004]
	Direct effect	–	.28	.14	2.06*	[.01, .55]
	Conditional indirect effect	Low (−1 SD)	.01	.03	.24	[−.05, .06]
		Mean	−.10	.05	−1.80	[−.20, .01]
		High (+1 SD)	−.20	.10	−1.98*	[−.40, −.002]
	Conditional total effect	Low (−1 SD)	.29	.14	2.06*	[.01, .56]
		Mean	.19	.11	1.71	[−.03, .40]
		High (+1 SD)	.08	.10	.86	[−.10, .27]
Positive Beliefs	IMM	–	−.04	.03	−1.17	[−.10, .03]
	Direct effect	–	.34	.09	3.58***	[.15, .52]
	Conditional indirect effect	Low (−1 SD)	.02	.02	1.33	[−.01, .06]
		Mean	−.01	.02	−.74	[−.05, .02]
		High (+1 SD)	−.05	.05	−1.04	[−.15, .05]
	Conditional total effect	Low (−1 SD)	.36	.10	3.59***	[.16, .56]
		Mean	.32	.09	3.45**	[.14, .51]
		High (+1 SD)	.29	.10	2.93**	[.09, .48]
Cognitive Confidence	IMM	–	−.06	.05	−1.30	[−.16, .03]
	Direct effect	–	.07	.11	.65	[−.15, .29]
	Conditional indirect effect	Low (−1 SD)	.01	.01	.89	[−.01, .04]
		Mean	−.05	.04	−1.23	[−.13, .03]
		High (+1 SD)	−.11	.09	−1.28	[−.29, .06]
	Conditional total effect	Low (−1 SD)	.08	.12	.72	[−.15, .31]
		Mean	.02	.09	.24	[−.15, .19]
		High (+1 SD)	−.04	.08	−.53	[−.20, .11]
Cognitive Self‐Consciousness	IMM	–	−.08	.05	−1.53	[−.17, .02]
	Direct effect	–	.40	.10	3.90***	[.20, .61]
	Conditional indirect effect	Low (−1 SD)	.04	.02	1.73	[−.01, .08]
		Mean	−.04	.03	−1.16	[−.10, .03]
		High (+1 SD)	−.11	.08	−1.41	[−.27, .05]
	Conditional total effect	Low (−1 SD)	.44	.11	3.87***	[.22, .66]
		Mean	.37	.10	3.61***	[.17, .56]
		High (+1 SD)	.29	.11	2.61**	[.07, .51]
Need for Control	IMM	–	−.07	.04	−1.86	[−.15, .004]
	Direct effect	–	.18	.12	1.54	[−.05, .41]
	Conditional indirect effect	Low (−1 SD)	.04	.03	1.55	[−.01, .10]
		Mean	−.03	.03	−1.22	[−.08, .02]
		High (+1 SD)	−.11	.06	−1.72	[−.23, .01]
	Conditional total effect	Low (−1 SD)	.22	.14	1.66	[−.04, .49]
		Mean	.15	.11	1.36	[−.07, .37]
		High (+1 SD)	.08	.10	.79	[−.11, .26]

Note: Values represent unstandardized regression coefficients (b), standard errors (SE), t values and 95% confidence intervals (CIs) from conditional process analyses. Conditional effects are estimated at ED = low (−1 SD), mean, and high (+1 SD).

Abbreviations: ED, emotional dysregulation; IMM, index of moderated mediation; MB, metacognitive beliefs.

^* p < .05;

^** p < .01;

^*** p < .001.

Negative beliefs

As shown in Table 4, a significant moderated mediation effect was found (b = −.10, SE = .05, t = −2.03, p = .042, 95% CI [−.20, −.004]). This indicates that ED significantly moderated the indirect effect of negative beliefs on academic performance via anxiety. Specifically, higher ED heightened the negative impact of negative beliefs on academic performance through anxiety. Figure 2 presents the moderated mediation model and standardized path coefficients.

FIGURE 2.

Moderated mediation model of the interaction between negative beliefs and ED in relation to academic performance through anxiety. ED, emotional dysregulation. Path coefficients are standardized. Age and Sex (0 = male, 1 = female) are included as covariates. *p < .05; **p < .01; ***p < .001.

Conditional indirect effects (Table 4) were computed to examine the indirect effect of negative beliefs on academic performance at different levels of ED. No significant indirect effect was found at low levels of ED (−1 SD) or at the mean level of ED (ps > .05). However, at high levels of ED (+1 SD), a significant negative indirect effect emerged (b = −.20, SE = .10, t = −1.98, p = .048, 95% CI [−.40, −.002]). Notably, there was a significant positive direct effect of negative beliefs on academic performance (b = .28, SE = .14, t = 2.06, p = .039, 95% CI [.01, .55]). The total effect of negative beliefs on academic performance was significant at the low ED level (b = .29, SE = .14, t = 2.06, p = .040, 95% CI [.01, .56]), but not significant at the mean and high ED levels (ps > .05).

Cognitive confidence, need for control, positive beliefs, cognitive self‐consciousness

As shown in Table 4, we did not find significant moderated mediation effects of ED on the relationships between other MB dimensions (cognitive confidence, need for control, positive beliefs, cognitive self‐consciousness) and academic performance via anxiety (ps > .05). Moreover, the indirect effects of these MB dimensions on academic performance via anxiety were not significant at any level of ED (ps > .05) (Table 4).

Nevertheless, positive beliefs had a significant direct effect on academic performance (b = .34, SE = .09, t = 3.58, p < .001, 95% CI [.15, .52]). They also showed a significant conditional total effect on any ED level: b = .36, SE = .10, t = 3.59, p < .001 at the low ED level; b = .32, SE = .09, t = 3.45, p = .001 at the mean ED level and b = .29, SE = .10, t = 2.93, p = .003 at the high ED level. Similarly, cognitive self‐consciousness had a significant direct effect on academic performance (b = .40, SE = .10, t = 3.90, p < .001, 95% CI [.20, .61]) and had significant conditional total effects: b = .44, SE = .11, t = 3.87, p < .001 at the low ED level; b = .37, SE = .10, t = 3.61, p < .001 at the mean ED level and b = .29, SE = .11, t = 2.61, p = .009 at the high ED level.

DISCUSSION

This study investigated the moderating role of ED in the mediated pathway from MB to academic performance via anxiety among Chinese adolescents. Two findings stand out. First, we found that ED moderated the relationship between negatively valenced MB and anxiety. Second, ED moderated the indirect effect of negative beliefs on academic performance through anxiety, with this detrimental pathway significant only at the high ED level. These findings highlight the importance of considering individual differences in emotion regulation when assessing the cognitive‐affective pathways that impact academic performance.

The relationship between MB, ED, anxiety and academic performance

Consistent with Hypothesis 1 and previous studies, all MB dimensions were positively correlated with ED and anxiety, with negative beliefs about worry showing relatively stronger associations. This pattern converges with earlier research (Ramos‐Cejudo & Salguero, 2017; Reinholdt‐Dunne et al., 2019) and recent meta‐analytic evidence in youth (Thingbak et al., 2024). The MB–anxiety link observed in our findings, along with Zhang et al.'s (2023) study on a Chinese non‐clinical adolescent sample, reinforces the applicability of the S‐REF model beyond clinical contexts. Together, these correlations reinforce MB as a theoretically and practically meaningful target for anxiety reduction in both clinical and school settings.

Notably, we did not observe a significant correlation between anxiety and academic performance, which contrasts with the findings reported by Wu et al. (2025). However, our results are consistent with broader evidence showing that the emotionality component of test anxiety, which closely resembles the DASS‐21 Anxiety construct that captures physiological arousal, is typically not harmful to academic outcomes (e.g., Chin et al., 2017; Steinmayr et al., 2016). Methodologically, the observed‐score correlation may also underestimate the underlying association due to measurement error; in contrast, SEM models anxiety as a latent construct and estimates its unique association with academic performance while accounting for covariates and shared variance with other predictors. Accordingly, the association between anxiety and academic performance may be more evident in the multivariate SEM than in zero‐order correlations.

Moderation of ED on the relationship between negatively Valenced MB and anxiety

Consistent with Hypothesis 2, ED significantly positively moderated the positive relationship between negatively valenced MB—negative beliefs about worry, low cognitive confidence and the need to control thoughts—and anxiety. By contrast, positive beliefs and cognitive self‐consciousness showed no ED‐contingent association with anxiety. This asymmetry is theoretically coherent. Since ED reflects failures in executive control of action that sustain CAS according to the S‐REF framework, negatively valenced MB dimensions that are more directly linked to worrying, perceived uncontrollability, and deficient cognition should therefore be most sensitive to ED‐related loss of top–down regulation. ED may reinforce maladaptive processes such as perseverative thinking and suppression that further increase anxiety. In contrast, effective regulation potentially buffers these effects by restoring cognitive and emotional control. This aligns with literature linking executive dysfunction, such as inhibitory control deficits, to difficulties of terminating harmful thought intrusions (Koster et al., 2017; Mohammed et al., 2022). Thus, ED may act as a regulatory gate that enables or prevents MB from triggering emotional and behavioural malfunctions. The interactive role of MB and ED extends findings by prior research that mostly examined MB and ED as independent predictors of anxiety (e.g., Salguero et al., 2019).

However, our findings diverge from Cécillon et al. (2024), who found no MB × ED interaction predicting trait anxiety. This discrepancy may be due to differences in anxiety measurement: trait anxiety reflects a stable disposition less sensitive to dynamic regulation, whereas general non‐trait anxiety is context‐dependent under academic stress and may better capture ED and MB's interactions (Spielberger & Vagg, 1995). In addition, we analysed MB at the dimensional level rather than using the global composite, suggesting that negatively valenced MB dimensions are more strongly implicated in interacting with ED to relate to anxiety. This underscores the importance of disaggregating MB to capture their nuanced effects on psychological outcomes in future studies.

Notably, ED amplified anxiety's associations with negatively valenced MB but not with positively valenced MB (Wells, 2000; Wells & Cartwright‐Hatton, 2004). Negatively valenced MB fosters excessive attention to emotional threat and promotes maladaptive coping strategies such as perseverative thinking, suppression and catastrophizing—processes amplified by high ED (Mansueto et al., 2022; Salguero & Ramos‐Cejudo, 2023). Perseverative thinking, which plays a central role in contributing to CAS, may particularly underlie this process (Palmieri et al., 2018; Weber & Exner, 2013). In contrast, positively valenced MB, which is less associated with anxiety and perseverative thinking (Spada et al., 2021; Thingbak et al., 2024), did not moderate in the current study. This dimensional specificity indicates that ED exacerbates anxiety primarily when coupled with MB dimensions, fostering rigid, ruminative coping. The exclusive moderation pattern observed among negatively valenced MB further underscores the importance of examining MB dimensions separately in future research.

The interaction effect of ED and negative beliefs, which is the MB dimension most consistently linked to anxiety across models and interventions (Clauss et al., 2019; Johnson & Hoffart, 2018), was particularly salient. Negative beliefs were positively associated with anxiety at both moderate and high ED levels, while the positive association was not found at the low ED level, suggesting that adaptive emotion regulation potentially buffers the maladaptive MB effects. This finding aligns with prior work showing that negative beliefs are especially pathogenic when combined with both chronic and acute regulatory deficits (Clauss et al., 2019; Gerges et al., 2023).

Interestingly, though typically viewed as maladaptive, the need for control dimension was negatively associated with anxiety among adolescents with low ED, resonating with calls to evaluate this dimension independently (Nieto et al., 2010). Need for control involves beliefs about suppressing dangerous thoughts to prevent harm (Tecuta et al., 2022; Wells & Matthews, 1996). Although previously linked positively to perseverative thinking (Spada et al., 2021), when coupled with effective regulation, it may instead induce goal‐directed vigilance and proactive coping rather than maladaptive rumination. This interpretation aligns with the theoretical claim of the S‐REF model: the expression and consequences of MB could depend on the ability and strategies of control, such that MB can function adaptively under adequate regulatory control (Wells, 2009).

Moderation of ED on the indirect effect of negative beliefs on academic performance via anxiety

Supporting Hypothesis 3, ED positively moderated the negative indirect effect of negative beliefs on academic performance through elevated anxiety. This detrimental pathway was significant only at the high ED level, and no other MB dimensions showed a significant moderated mediation effect. This underscores the unique role of negative beliefs in channelling dysfunctional MB into academic underachievement. This mediation pattern aligns with Huntley et al.'s (2023) findings, which indicate that negative beliefs, but not other MB dimensions, were indirectly associated with poorer academic performance through anxiety (Huntley et al., 2023). Our study extends the literature in two aspects: we focus on general non‐trait anxiety as the mediator, and we identify ED as a boundary condition that determines when MB indirectly harms academic performance via anxiety.

The absence of a negative indirect effect at the low ED level suggests a potential protective role for adaptive emotion regulation. This pattern may reveal one possible mechanism underlying the extensive literature demonstrating that ED impairs adaptive functioning, including cognitive development, task management and academic performance (e.g., Garber & Dodge, 1991; Graziano et al., 2007; Kamphaus, 1987). Our findings add precision by locating anxiety as the key mechanism in this process: when ED is low, adolescents may contain anxiety and thereby potentially buffer the academic underachievement associated with maladaptive MB.

Notably, we also observed a positive direct effect of negative beliefs on academic performance in the moderated mediation model, which counterbalanced the negative indirect pathway at the high level of ED. As a result, negative beliefs showed a positive total effect on performance when ED was low and did not produce an overall negative effect when ED was high. This pattern showed an inconsistent mediation (suppression) effect where the direct effects and the indirect effects have opposed directions, and the total effect approaches zero compared to the direct effect (MacKinnon et al., 2000). This pattern also diverges from findings based on Western samples, where negative beliefs were negatively linked to academic performance (Cécillon et al., 2024), and negative beliefs did not predict academic performance when test anxiety was treated as a mediator (Huntley et al., 2023).

Nevertheless, the discrepancy may be attributed to culture. In Confucian‐heritage, high‐stakes educational systems, worry is often harnessed as performance fuel despite psychological costs (Ji, 2010). Within this context, negative beliefs about worry—understood as heightened concern about the danger of unchecked worry—may prompt vigilance and self‐regulatory tightening, such as earlier planning and over‐preparation in the educational context, especially given that school‐related stressors dominate adolescents' concerns (Cheng & Lin, 2023; Sun et al., 2013). Nevertheless, this does not imply that stronger negative beliefs are inherently adaptive; rather, under specific cultural and regulatory conditions, their behavioural correlates can coincide with achievement. Future research should examine the long‐term academic effects of negative beliefs across cultures, assessing whether the short‐term gains observed among Chinese adolescents are offset by deeper, longer‐term costs.

Taken together, our findings support the applicability of the S‐REF model among Chinese adolescents in mainstream educational settings. While previous studies have linked ED to poor academic outcomes via deficits in self‐efficacy (Usán Supervía & Quílez Robres, 2021), executive dysfunction (Soltani Nezhad & Delroba, 2024) or behavioural problems (Graziano et al., 2007), we offer an integrative account grounded in the S‐REF framework. Specifically, we show that ED exacerbates the negative indirect effects of negatively valenced MB on academic performance by potentially maintaining high anxiety levels and perpetuating a cycle of affective–cognitive dysfunction.

Positive beliefs, cognitive self‐consciousness and academic performance

An unexpected additional finding was the positive overall effect of positively valenced MB, including cognitive self‐consciousness and positive beliefs, on academic performance. Although previous research frequently reported negative associations between MB and academic outcomes (e.g., Cécillon et al., 2024; Huntley et al., 2023), the present findings suggest that these relationships may be context‐dependent and reflect the adaptive facets of MB.

Cognitive self‐consciousness, defined as the tendency to monitor one's own thinking (Marker et al., 2006; Wells & Cartwright‐Hatton, 2004), conceptually overlaps with metacognitive awareness, a construct linked to improved learning and academic functioning (Abdelrahman, 2020; Tuononen et al., 2023). In high‐pressure academic environments, such as Chinese middle schools, this strong self‐monitoring may facilitate the development of effective study strategies, goal‐setting and planning (De Boer et al., 2018; Hayat et al., 2020). Nevertheless, excessive self‐focused monitoring may become counterproductive if it escalates into rumination among clinical populations; therefore, the positive overall effect should be interpreted with caution according to the S‐REF model.

Similarly, positive beliefs about worry—the belief that worry is useful or protective (Dugas & Koerner, 2005; Wells, 2005)—are also positively linked to academic performance. This suggests that although such beliefs are implicated in the maintenance of pathological worry, they may also serve a motivational function in performance‐driven settings. Within the Chinese educational and Confucian cultural context, where academic achievement is highly valued and perseverance is socially reinforced, positive beliefs about worry may channel anxiety into disciplined goal pursuit (Borkovec & Roemer, 1995; Liu et al., 2020). These findings underscore the importance of exploring the specific effects of different dimensions of MB and highlight the potential cultural differences in examining the emotional and behavioural outcomes of MB.

Limitations

This study has several limitations. First, data were collected via convenience sampling at a single middle school in Zhejiang Province, which may have introduced selection bias and limited the generalizability of the findings to Chinese adolescents. Additionally, only participants' biological sex was recorded, and gender identity was not included as a covariate in the analyses. Future studies should include more diverse demographic variables, recruit samples from diverse regions and socioeconomic strata and consider multilevel modelling to examine contextual variability and the robustness of the findings. Second, the cross‐sectional design precludes causal inference regarding the relationships among MB, ED, anxiety and academic performance. Future research should replicate these effects in larger samples using longitudinal or intervention methods, such as incorporating metacognitive therapy, to facilitate validity and generalizability. Such replication would provide a stronger empirical foundation for the theory development. Third, reliance on self‐report measures may have introduced common‐method bias, including social desirability bias and recall inaccuracies (Wright, 2005). To enhance validity, future research should adopt multi‐method approaches, such as physiological indicators or informant ratings (Tichon et al., 2014).

Future work should also explore: (1) incorporating both test anxiety (worry and emotionality) and general non‐trait anxiety, along with other school‐related anxiety domains, to more comprehensively examine their unique and overlapping associations with academic performance; (2) how MB interacts with executive functioning across different mental health conditions beyond anxiety, such as depression and obsessive‐compulsive disorder (Solem et al., 2009; Tümkaya et al., 2018), which share common metacognitive vulnerabilities; (3) the mechanisms through which adaptive emotion regulation buffers MB's negative influences, such as through reduced perseverative thinking or improved cognitive control and (4) the design and evaluation of integrated interventions that target both MB and ED to disrupt the cycle of cognitive–behavioural dysfunction. Finally, longitudinal and cross‐cultural research is especially needed to replicate the present findings and trace the long‐term developmental trajectories of MB and ED, thereby identifying critical periods for effective intervention.

Implications

Grounded in the S‐REF model, this study provides new insights into the cognitive–emotional dynamics underlying adolescent anxiety and academic performance. The findings provide theoretical, practical and policy implications for school‐based anxiety reduction.

From a theoretical perspective, this study extends the S‐REF framework to a Chinese non‐clinical adolescent sample, demonstrating its applicability beyond clinical samples. The dimensional specificity of the findings underscores the value of examining MB dimensions separately. Furthermore, the pattern of findings suggests cultural differences in how specific MB dimensions link to performance, reflecting the potential influence of Confucian norms that emphasize self‐discipline, perseverance and self‐control.

From an applied perspective, MB and ED may serve as practical components in school‐based screening to identify students at risk of anxiety and academic difficulties. Interventions such as attention training, detached mindfulness (Haseth et al., 2019; Wells, 2009) and cognitive reappraisal (Jamieson et al., 2010; Johns et al., 2008), which are derived from metacognitive therapy, may enhance cognitive flexibility and emotional control. Evidence‐based programs, such as Emotion Regulation Therapy (Renna et al., 2017), Emotion‐Focused CBT (Suveg et al., 2018) and digital interventions (Emmelkamp et al., 2022; Wisman et al., 2023), have demonstrated effectiveness in enhancing emotional resilience. Integrated interventions such as Meta‐Emotion Therapy for Adolescents (Cain et al., 2024) also warrant further evaluation.

These implications are especially relevant in high‐pressure environments like Chinese middle schools, where academic stress and emotional suppression are common. Therefore, schools should implement culturally responsive strategies by reducing stigma, training educators to identify anxiety among students and integrating adapted social–emotional learning curricula that foster openness, regulation and resilience.

CONCLUSION

This study highlights the interactive roles of MB and ED in shaping adolescents' academic performance via anxiety. While all MB dimensions were linked to greater anxiety, ED moderated the relationship between anxiety and negatively valenced MB dimensions. Among all MB dimensions, only negative beliefs showed a significant indirect effect on academic performance, although it also had a positive direct effect and a conditional total effect. These findings extend the S‐REF model to Chinese educational contexts and emphasize the need to examine MB dimensions separately. Future research should explore how cognitive and emotional vulnerabilities jointly influence learning outcomes across diverse cultures and student populations and how targeted interventions might disrupt these maladaptive pathways.

AUTHOR CONTRIBUTIONS

Zeyiting Gao: Conceptualization; writing – original draft; methodology; validation; visualization; writing – review and editing; software; formal analysis; project administration; data curation; resources. Yuran Chen: Writing – review and editing; resources; validation. Chun Chen: Writing – review and editing; project administration; supervision.

CONFLICT OF INTEREST STATEMENT

The authors report there are no competing interests to declare.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are openly available in OSF at https://osf.io/5fdzn/?view_only=e4d9cf229dc841dd9d66a088068419fc.