The target population of this study is the Chinese undergraduate students studying in normal universities of H province. The sample size should be determined based on the literature and statistical methods used in the study. Considering that the data analysis method used in this study includes CFA, the current study referred to the sample size determination proposed by (Wolf et al. 2013). According to these criteria, the minimum sample size required for a CFA model with three factors and three indicators for each factor should be about 420 to get the factor loadings of 0.5 and this sample size can meet all the priori criteria in the literature. Under this guideline, 515 participants were recruited from three normal universities in H province, who can meet the criteria including no serious physical illness, no diagnosed psychological problems or psychiatric illness.

A total of 498 valid questionnaires were obtained and the age of the participants ranged from 17 to 24 (M = 19.15 ± 1.03; 66.1% females). Descriptive analysis was conducted based on the demographic information of respondents, which reflects the distribution of respondents on different demographic variables. The results show that there were 140 (28.1%) undergraduate students came from the first academic year, and 127 (25.5%), 116 (23.3%), and 115 (23.1%) respondents from the second to the fourth academic year, respectively. Among the participants, 218 (43.8%) were majoring in humanities, 156 (31.3%) participants in science and Engineering, and 124 (24.9%) in arts.

This study employed the Chinese version of the Utrecht Work Engagement Scale-Student (UWES-S) and its short version (UWES-S-9) to assess academic engagement among undergraduate students. Developed by (Schaufeli et al. 2002a), the UWES-S comprises three subscales: Vigor, Dedication, and Absorption. The 17-item scale utilizes a seven-point Likert format, with responses ranging from 0 (“never”) to 6 (“always/every day”). Higher scores indicate greater academic engagement. In this study, the Chinese version adapted by (Li and Huang 2010) was referred to. The subscales demonstrated good composite reliability with values of 0.838 (Vigor), 0.820 (Dedication), and 0.812 (Absorption) based on the current sample.

The UWES-S-9, a 9-item adaptation of the UWES-9, was tailored for educational contexts (Schaufeli et al., 2006). It maintains the three-subscale structure, with each subscale containing three items. Vigor assesses energy and enthusiasm in learning, Dedication reflects positive emotional experiences from learning, and Absorption measures concentration levels during the study. The Chinese version of the UWES-S-9 used in this study was likewise translated by (Li and Huang 2010).

Academic burnout was evaluated using the Learning Burnout Scale of Undergraduates (LBS), adapted by (Lian et al. 2005) based on the Maslach Burnout Inventory-Student Survey (MBI-SS) (Schaufeli et al., 2002b). This 20-item scale encompasses three dimensions: Low Mood, Inappropriate Behavior, and Low Sense of Accomplishment. Items are rated on a five-point Likert scale from 1 (“Not at all like me”) to 5 (“Very much like me”), yielding total scores between 20 and 100. Higher scores indicate greater academic burnout. The study found good reliability for the LBS subscales, with composite reliability values of 0.831 (Low Mood), 0.785 (Inappropriate Behavior), and 0.756 (Low Sense of Accomplishment).

This research received ethical approval from the Ethics Committee for Research involving Human Subjects of University Putra Malaysia Research (NO: JKEUPM-2023-137). After that, a pilot study (n=203) was conducted to check if there was any misunderstanding among the participants toward the items. It was found that all the items can be understood clearly and there was no ambiguous expression. Two weeks later, to evaluate the test-retest reliability of the questionnaire, the respondents who participated in the pilot study were retested (n=203). Finally, the electronic version of the questionnaire was distributed to the participants through the Questionnaire Star platform and 515 respondents participated in the actual study.

After data collection, SPSS 26.0 was used for data examination, descriptive statistical analysis, and reliability tests. Subsequently, multiple competing measurement models were systematically examined using Mplus 8.3 to determine the optimal factorial representation of the UWES-S-9.

The analysis proceeded through four distinct phases: (1) confirmatory factor analysis (CFA) comparing one-factor and three-factor models, (2) exploratory structural equation modeling (ESEM) to address potential model misspecification, (3) bifactor-CFA incorporating both general and specific factors, and (4) bifactor-ESEM combining the advantages of both bifactor modeling and ESEM approaches.

The one-factor model specified all nine items loading onto a single latent construct representing overall academic engagement. The three-factor model posited items loading onto three correlated dimensions—Vigor, Dedication, and Absorption—consistent with previous theoretical expectations. The ESEM models allowed for cross-loadings between factors while maintaining the target factor structure. Bifactor models incorporated both a general engagement factor and three specific factors (Vigor, Dedication, Absorption) to examine the extent to which items reflected general vs. specific variance components.

Model adequacy was assessed using multiple complementary approaches. Conventional goodness-of-fit indices included the chi-square to degrees of freedom ratio (χ²/df), root mean square error of approximation (RMSEA) with 90% confidence intervals, comparative fit index (CFI), Tucker-Lewis index (TLI), and standardized root mean square residual (SRMR). Acceptable model fit was indicated by CFI and TLI values ≥ 0.90, RMSEA ≤ 0.08, χ²/df ≤ 5.0, and SRMR ≤ 0.08 (Hu and Bentler, 1999; Tabachnick et al., 2013). Model comparison was further informed by information criteria, specifically the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), with lower values indicating superior balance between goodness of fit and parsimony (Vrieze, 2012).

For bifactor models, additional complementary indices were examined to evaluate the appropriateness of the bifactor solution. These included explained common variance (ECV), percentage of uncontaminated correlations (PUC), omega hierarchical coefficients for the general factor (ωH), and omega hierarchical coefficients for specific factors (ωHS). ECV values above 0.70 suggest that the general factor accounts for most common variance, while PUC values above 0.80 are generally preferred, though moderate values (0.50–0.70) can be acceptable in models with multiple specific factors (Rios and Wells, 2014). Omega hierarchical coefficients (ωH) above 0.70 for the general factor support reliable interpretation of total scores, while low ωHS values (< 0.30) indicate limited reliable variance in specific factors beyond the general factor (Smits et al., 2015).

Internal consistency was evaluated using multiple indices appropriate for different factorial structures. McDonald's omega coefficients were computed as more appropriate reliability estimates for multifactor structures, with the threshold of ≥ 0.70 (Hayes and Coutts, 2020). For bifactor models, omega hierarchical coefficients were calculated separately for general and specific factors to assess the reliability of total scores vs. subscale scores. Composite reliability (CR) was also examined, with values ≥ 0.60 considered adequate (Fornell and Larcker, 1981). Corrected item-total correlations above 0.30 were considered indicative of adequate item contribution (Artanti et al., 2018). Test-retest reliability was evaluated using intraclass correlation coefficients (ICCs), with values ≥ 0.75 reflecting good stability and values between 0.50 and 0.75 indicating moderate stability (Koo and Li, 2016).

Factor distinctiveness between UWES-S-9 dimensions was evaluated through examination of factor correlations, average variance extracted (AVE), and shared variance between factors (Fornell and Larcker, 1981). Standard factor loadings were required to exceed 0.50 and remain below 1.0 (Hair et al., 2010). Factor distinctiveness validity was supported when the AVE of each factor exceeded the shared variance with other factors, indicating that factors capture more unique variance than shared variance with other constructs.

Maximum Likelihood Estimation was used in the analysis since the collected data met the normality assumptions of parametric tests (van Zyl and ten Klooster, 2022). The correlation relationship between UWES-S-9 and its original scale (UWES-S) as well as its relationship with academic burnout was tested through bivariate correlations. A sequential multigroup CFA approach was implemented following established methodological procedures to assess the measurement invariance of the UWES-S-9 across gender, academic year, and major (Chen, 2007; Kang et al., 2016). The measurement invariance assessment proceeded through three hierarchical levels of constraint. Initially, configural invariance was examined by simultaneously estimating the baseline model across gender, academic year, and major without imposing equality constraints, thereby testing whether the factor structure remained consistent across groups. Subsequently, metric invariance was assessed by constraining factor loadings to be equal across groups while allowing intercepts and residual variances to vary freely, examining whether the relationship between latent factors and observed indicators remained equivalent. Finally, scalar invariance was evaluated by imposing additional equality constraints on item intercepts across groups. Model comparison was conducted using chi-square difference tests alongside practical fit indices, with invariance supported when changes in model fit indices (ΔRMSEA, ΔSRMR, ΔCFI, ΔTLI, and ΔMNCI) did not exceed the established cutoff criteria of 0.015, 0.03, 0.01, 0.008, and 0.02, respectively (Chen, 2007; Kang et al., 2016).

Comprehensive data screening procedures were implemented to ensure data quality and statistical assumptions for confirmatory factor analysis. The online survey platform prevented incomplete submissions, resulting in no missing values. However, 15 cases were excluded based on cutoff values (Collier, 2020; Lanning, 1989; Tabachnick et al., 2013): 3 participants with insufficient response time (< 2 min), 2 cases with invariant response patterns (SD < 0.50); 4 univariate outliers (z-scores exceeding ±3.29), and 8 multivariate outliers (Mahalanobis d² with p1 and p2 < 0.001). The final sample comprised 498 participants (96.70% valid response rate). Normality assessment revealed skewness (0.138–1.052) and kurtosis (0.865–1.738) coefficients within acceptable ranges (Collier, 2020; Tabachnick et al., 2013), confirming distributional assumptions for subsequent analyses.

Chi-square difference tests revealed violations of tau-equivalence assumptions across all UWES-S-9 subscales [Δχ²₍₆₎ = 109.839, p < 0.001], indicating heterogeneous factor loadings. Consequently, internal consistency reliability was assessed using McDonald's omega coefficients derived from a congeneric three-factor CFA, as this approach demonstrates greater robustness to unequal factor loadings compared to coefficient alpha. The omega coefficients for Vigor, Dedication, and Absorption factors of UWES-S-9 were 0.800, 0.786, and 0.854, respectively, indicating adequate reliability. Additionally, the composite reliability (CR) values for the UWES-S-9 subscales—Vigor, Dedication, and Absorption—were 0.70, 0.79, and 0.87, respectively, reflecting acceptable levels of reliability. However, the traditional reliability estimates should be interpreted cautiously in light of the bifactor analysis results, which suggest that most reliable variance is attributable to the general factor rather than specific dimensions.

In addition to internal consistency, test–retest reliability was examined to evaluate the temporal stability of the UWES-S-9. Participants (n = 203) completed the scale at two time points with a 14-day interval. Results indicated moderate to good test-retest reliability for the overall UWES-S-9 [ICC = 0.771, 95% CI (0.709, 0.821)] and its subscales: Vigor [ICC = 0.745, 95% CI (0.677, 0.801)], Dedication [ICC = 0.778, 95% CI (0.718, 0.827)], and Absorption [ICC = 0.773, 95% CI (0.711, 0.823)]. These findings suggest that the UWES-S-9 demonstrates temporal stability, supporting its use as a reliable measure of academic engagement in educational settings.

As shown in Table 5, all item-corrected total correlations exceeded 0.50, providing additional descriptive support for internal consistency. Although Cronbach's alpha assumes tau-equivalence and may either underestimate or overestimate reliability when this assumption is violated, the UWES-S-9 demonstrated a high alpha coefficient in this study (α = 0.901), indicating strong internal consistency. “Alpha if item deleted” analyses also yielded consistently high values ranging from 0.886 to 0.898, suggesting that removing any single item would not improve reliability. These results, while interpreted with caution, are consistent with the omega-based findings and further suggest that the scale has good internal consistency.

Construct validity of a measurement model can be evaluated by checking whether the fit indices have achieved the recommended threshold. The comprehensive analytical approach provided robust evidence for construct validity through multiple complementary criteria. While traditional fit indices supported the adequacy of the three-factor structure, the superior performance of ESEM models and the bifactor analysis results provided deeper insights into the construct's dimensional nature. The bifactor-CFA model, despite showing excellent fit statistics, exhibited convergence problems with implausible parameter estimates, rendering it unsuitable for interpretation. Nevertheless, the bifactor-ESEM results provided compelling evidence that the optimal structure of the UWES-S-9 is unidimensional, as the general engagement factor accounted for 79% of the common variance and minimal reliable variance attributable to specific Vigor, Dedication, and Absorption dimensions. The good model fit and strong omega hierarchical coefficient for the general factor (ωH = 0.827) support the use of a total engagement score, while the negligible specific factor reliabilities suggest limited utility of subscale scores. Although the moderate PUC value (0.667) indicates that treating the UWES-S-9 as perfectly unidimensional may introduce some parameter bias, the pattern of omega coefficients strongly favors a unidimensional interpretation over the traditional three-factor structure. The convergence of evidence from multiple analytical approaches strengthens confidence in the scale's construct validity while clarifying the predominantly unidimensional nature of the engagement construct as measured by the UWES-S-9.

The results revealed a strong positive correlation between the UWES-S-9 and the original UWES-S-17 (r = 0.961; p < 0.01), indicating substantial agreement between the two versions. As detailed in Table 6, each subconstruct of the UWES-S-9 demonstrated robust correlations with its corresponding subconstruct in the original UWES-S-17. Specifically, the Vigor subconstructs showed a significant positive relationship (r = 0.750; p < 0.01), as did the Dedication subconstructs (r = 0.814; p < 0.01) and the Absorption subconstructs (r = 0.806; p < 0.01). These findings provide robust evidence of the scale equivalence with the UWES-S-17, supporting its capacity to measure the same theoretical constructs as the original scale in a more concise format.

To examine the relationship between academic engagement and academic burnout, a Pearson correlation analysis was performed. As illustrated in Table 7, the findings revealed significant negative correlations between academic engagement and academic burnout, including their respective subconstructs. Notably, a strong negative correlation was observed between overall academic engagement and academic burnout (r = −0.627; p < 0.01). Further analysis of the subconstructs showed varying degrees of negative correlations. The weakest, yet still significant, negative correlation was found between the Absorption subscale of UWES-S-9 and the Inappropriate Behavior subscale of LBS (r = −0.411; p < 0.01). Conversely, the strongest negative correlation was observed between the Vigor subscale of UWES-S-9 and the overall LBS score (r = −0.623; p < 0.01). These results suggest a negative relationship between academic engagement and academic burnout across various dimensions of these constructs, which are consistent with previous studies examining this relationship, providing substantial evidence for the concurrent validity of the UWES-S-9.

Multigroup CFA was conducted to assess the measurement invariance of the UWES-S-9 across gender, academic year, and major. Results as shown in Table 8 indicated that the changes in model fit indices of the ΔRMSEA, ΔSRMR, ΔCFI, Δγ^, and ΔMNCI between configural and metric, and metric and scalar were no more than the cutoff values of 0.015, 0.03, 0.01, 0.008, 0.02, respectively (Chen, 2007; Kang et al., 2016), supported configural, metric, and scalar invariance, indicating that the scale's structure, factor loadings, and item intercepts are equivalent across these demographic groups. This establishes the UWES-S-9's validity for meaningful comparisons of academic engagement levels between genders and across different academic years and majors.

This study employed a comprehensive analytical framework to examine the factorial structure of the UWES-S-9, extending beyond traditional CFA to include ESEM and bifactor modeling approaches. The systematic comparison revealed important insights into the dimensional nature of this instrument.

The traditional three-factor CFA model, while demonstrating acceptable fit indices, exhibited substantial interfactor correlations (ranging from 0.653 to 0.868) and discriminant validity concerns, particularly between the Vigor-Dedication factor pair (shared variance = 0.754). These findings align with previous observations of high intercorrelations among UWES dimensions (Carmona-Halty et al., 2019; Dominguez-Lara et al., 2025; Loscalzo and Giannini, 2019; Rastogi et al., 2018; Römer, 2016; Schaufeli et al., 2006), suggesting potential challenges with the theoretical distinctiveness of the three proposed factors. The ESEM approach partially addressed these issues by allowing cross-loadings between factors, resulting in improved discriminant validity properties and superior model fit, which aligns with previous findings (Dominguez-Lara et al., 2025). The three-factor ESEM model provided superior fit statistics and more favorable factor distinctiveness properties, with reduced interfactor correlations and improved parameter estimates. This improvement suggests that allowing for cross-loadings between factors better accommodates the complex structure underlying academic engagement, consistent with recent recommendations for ESEM applications in psychological measurement (Marsh et al., 2014).

However, the bifactor analysis provided the most informative perspective on the scale's dimensional structure. While the bifactor-CFA model exhibited convergence issues (evidenced by implausible parameter estimates), which aligns with the previous study (Dominguez-Lara et al., 2025), the bifactor-ESEM model revealed theoretically coherent results. The general factor accounted for 79% of common variance, with a substantial omega hierarchical coefficient (ωH = 0.827), indicating strong evidence for a unidimensional interpretation. In contrast, the specific factors exhibited minimal unique reliable variance (ωHS ≤ 0.149), indicating that the subscale scores offer limited additional value beyond the general engagement factor. This finding aligns with the conclusions of prior studies conducted in Ecuador (Dominguez-Lara et al., 2025).

The moderate percentage of uncontaminated correlations (PUC = 0.667) indicates that while treating the scale as perfectly unidimensional may introduce some parameter bias, the evidence strongly favors a general engagement interpretation over distinct subscale usage. These findings diverge from some validation studies that have supported distinct three-factor structures (Meng and Jin, 2017; Rastogi et al., 2018; Sulla et al., 2023) while aligning with research suggesting predominantly unidimensional conceptualizations of academic engagement (Chi et al., 2023; Dominguez-Lara et al., 2025; Loscalzo and Giannini, 2019; Römer, 2016; Serrano et al., 2019). The inconsistency in reported factor structures across studies may be attributed to variations in sample characteristics, statistical methodologies, and cultural contexts, though these potential explanations require further investigation.

The second research objective of this study is to assess the internal consistency reliability, test-retest reliability, composite reliability, construct validity, its scale equivalence and concurrent validity of UWES-S-9. Findings indicate that the UWES-S-9 exhibit good internal consistency reliability, test-retest reliability, and composite reliability as well as the construct validity. This study found higher test-retest reliability coefficients for the UWES-S-9 compared to previous validation studies of UWES-S conducted in Australia and Norway (Schaufeli et al., 2006). This discrepancy may be attributed to differences in the time interval between test administrations. The current study employed a 2-week interval, while the previous study used a 1-year interval. This methodological difference underscores the importance of considering temporal factors when interpreting stability coefficients, as shorter intervals typically yield higher reliability estimates due to reduced genuine change and enhanced memory effects.

Besides that, the values of Composite Reliability (CR) for the three sub-constructs of UWES-S-9 in this study align with findings from previous research conducted in Italy (Loscalzo and Giannini, 2019; Sulla et al., 2023), India (Rastogi et al., 2018), and Spain (Serrano et al., 2019). These consistent results across different cultural contexts provide strong evidence for the scale's reliability and validity in cross-cultural research. This comparison of psychometric properties across studies and cultures underscores the robustness of the UWES-S-9 as a measure of academic engagement. However, the bifactor analysis provided critical insights that contextualize these traditional reliability estimates. The omega hierarchical coefficient for the general factor (ωH = 0.827) substantially exceeded the threshold for reliable interpretation, while specific factor reliabilities were negligible (ωHS ≤ 0.149). This pattern suggests that while the scale demonstrates overall reliability, the majority of reliable variance is attributable to a general engagement factor rather than specific dimensional components.

This study revealed a significant correlation between the UWES-S-9 and its original scale, the UWES-S, supporting its scale equivalence. This finding aligns with previous validation research conducted across multiple countries, including Australia, Belgium, Canada, Finland, France, Germany, Netherlands, Norway, South Africa, and Spain, utilizing a comprehensive international database (Schaufeli et al., 2006). The consistency of these results across diverse cultural contexts provides strong evidence that the UWES-S-9 effectively measures the same underlying constructs as the original, longer scale. This supports the scale equivalence of the UWES-S-9 as a more concise instrument for assessing academic engagement, which also enhances our understanding of the UWES-S-9's applicability and reliability across a broader range of cultural settings, reinforcing its utility as a globally relevant measure of academic engagement.

The relationship between academic engagement and academic burnout was examined in this study and the results revealed significant negative correlations between academic engagement and academic burnout, including their respective subconstructs, which is in line with the previous validation studies conducted in Korea, Italy, Australia, Belgium, Canada, Finland, France, Germany, Netherlands, Norway, South Africa, and Spain (Rastogi et al., 2018; Römer, 2016; Schaufeli et al., 2006). The cross-cultural stability of the negative correlation between academic engagement and academic burnout further verifies the concurrent validity of the UWES-S-9.

To assess measurement invariance across demographic variables, multi-group CFA was employed. The findings indicated that the UWES-S-9 demonstrates consistent measurement across various demographic groups. Specifically, the analysis supported measurement invariance for the UWES-S-9 when considering gender, academic year, and field of study. These results align with previous research findings, such as those reported by (Chi et al. 2023) and (Carmona-Halty et al. 2019). This concordance with established literature lends further support to the robustness of the present outcomes. The observed invariance suggests that the UWES-S-9 maintains its psychometric properties and measures the intended construct equivalently across these diverse demographic categories.