Evaluating the factor structure, reliability and validity of the Copenhagen Burnout Inventory-Student Survey (CBI-SS) among faculty of arts students of Ekiti State University, Ado-Ekiti, Nigeria

The Copenhagen burnout inventory-student survey (CBI-SS) has shown promising psychometric properties in diverse student populations. This study aims to investigate the psychometric properties of the Nigerian version of the CBI-SS. This was a cross-sectional study of 635 students from Ekiti State University, Ado-Ekiti, Nigeria. Confirmatory factor analysis (CFA) was utilized to assess the CBI-SS validity. The reliability score of the CBI-SS was 0.957, ranging from 0.862 to 0.914 for the subscales. Correlation coefficients among the four CBI-SS factors ranged from 0.507 to 0.713. The CFA indicated an adequate goodness-of-fit for the four-factor model of the CBI-SS with the sample data. However, Item 10 was removed due to unacceptably low Average Variance Extracted score. The four factors demonstrated a negative correlation with both General Academic Self-Efficacy Scale and Cumulative Grade Point Average. Furthermore, both self-reported burnout and perceived course stress showed associations with the CBI-SS, where lower levels of burnout corresponded with lower median scores on the CBI-SS scales. This study underscores the significance of the CBI-SS in evaluating student burnout within our student population. The findings indicate that the CBI-SS is a highly reliable and valid instrument for assessing student burnout, suggesting its potential for effective utilization in the Nigerian academic context.


Study instruments
The study instrument consisted of three sections: Section One: This section included socio-demographic and academic-related items such as age, sex, undergraduate course, level, financial challenges, ill-health, CGPA, thoughts about quitting the course, perceived course stress, and self-reported burnout.
Section Two: This section contains the Copenhagen Burnout Inventory Student Version (CBI-SS), adapted from the Campos et.al modification of the original CBI for use among students 11 .This version of the CBI consists of a total of twenty-five questions across four domains, as follows: study-related burnout (7 items), and six items each for personal burnout, colleague-related burnout, and teacher-related burnout.
We modified the questionnaire to ensure clarity and relevance for our respondents.For example, all instances of "colleagues" were replaced with "students" and "teachers" with "lecturers", and factor 3 was renamed 'Fellow Student Related Burnout' (FSRB) from 'Colleague Related Burnout.' Finally, the questionnaire items were reworded to make them suitable for agreement-based Likert-type responses: Strongly Agree (5), Agree (4), Neutral (3) Disagree (2), Strongly Disagree (1).For example, item 1 was changed from "How often do you feel

Data analysis
The data downloaded from Google Form in Microsoft Excel format, were imported into IBM SPSS Version 25 for initial cleaning, in preparation for Confirmatory Factor Analysis (CFA) using Structural Equation Modeling (SEM) in JASP Version 0.18.1.0 25,26.
The data were prepared for CFA by examining for missing data, outliers and unengaged responses [27][28][29] .Unengaged responses are those records in which respondents have filled the survey form without paying attention to the questions and answering them consistently with the same number 27,29 .Thirty-nine (5.8%) of the 674 respondents were either unengaged or had missing data greater > 60% and were excluded from further analysis, leaving 635 respondents for the final analysis.For those with less than 60% missing data, missing responses were replaced with the medians of their two nearest neighbors.Both the Cronbach's Alpha and the McDonald's Omega (ω), with a minimum acceptable value of 0.70, were employed to evaluate the reliability of the model and its constituent factors 30 .Despite its limitations, we used Cronbach's Alpha due to its widespread recognition and historical precedent in measuring reliability.It allowed for comparison with other studies.McDonald's Omega was employed due to its capacity to offer a more precise assessment of internal consistency when multidimensional constructs are present.By employing this dual approach, a thorough evaluation of reliability is achieved, as it merges the familiarity of Cronbach's Alpha with the more nuanced perspectives provided by McDonald's Omega 31 .Normality of the items was checked using Kolmogorov-Smirnov procedure.
CFA was conducted to verify if the factor structure proposed by Campos et al. in Brazil presented an adequate fit for this study's sample 11 .Upon conducting the Kolmogorov-Smirnov test, it was determined that all 25 items in the dataset were not normally distributed; consequently, the Diagonally Weighted Least Squares (DWLS) method, coupled with robust error calculation, was employed as the estimator for the CFA to accommodate the ordinal nature and distribution characteristics of the data [32][33][34] 35 .The Heterotrait-Monotrait (HTMT) ratio assessed discriminant validity, while Average Variance Extracted (AVE) evaluated convergent validity.Adequate convergent and discriminant validity were indicated by AVE > 0.50 and HTMT < 0.90 35,36 .When the model did not fit, factor loadings, modification indices, and residuals were examined for model respecification 28,29 .The instrument's concurrent validity was evaluated by correlating it with CGPA and GASE, as well as by testing its association with self-reported burnout and stress.

Ethical approval
An institutional ethical approval was obtained from the Ekiti State University Teaching Hospital Institutional Review Board, which approved the study protocol.The online survey form contained a detailed introductory section providing information about the study's purpose, procedures, potential risks and benefits, confidentiality, and the participants' right to withdraw from the study at any time.All study participants gave informed consent and voluntarily agreed to participate in the research, and the study was conducted in accordance with the relevant guidelines and regulations.

Result
The validation process of the CBI-SS involved a total of 635 undergraduates from the seven departments of the Faculty of Arts (FAT) at Ekiti State University, Ado, South-West Nigeria.The demographic characteristics of the respondents from the seven departments are detailed in Table 2.The ratio of male to female respondents is 1:1.99,indicating that the proportion of female students participating in the survey was significantly higher than that of male students when compared to the overall student population of the faculty (p=0.015).

Construct validity
The Burnout.However, the factor loading for Item 10 of the Student-Related Burnout (SRB) dimension was only 0.123, falling below the recommended threshold of 0.5 for acceptable factor loading.Additionally, the residual variance for this item was above 1.0, suggesting that a substantial portion of its variance is not explained by its associated factor.The SRB latent factor of this item exhibited an AVE of 0.455, below the recommended minimum of 0.5.As a result of these two observations, we removed item 10, improving the AVE of SRB factor to 0.524.This result indicated that the convergent validity of the model was enhanced by eliminating SRB item 10.The HTMT ratios for the model varied between 0.561 and 0.818, values that are lower than 0.90, indicating satisfactory discriminant validity among the factors.These findings suggest that each factor in the model is sufficiently distinct, supporting the theoretical framework of the study.

Correlation
As demonstrated in Table 3, the Spearman correlation coefficients among the four CBI-SS factors range from 0.507 (between LRB and PB) to 0.713 (between FSRB and LRB).This range suggests that these four factors are significantly related to constructs of burnout.Additionally, Figure 1 illustrates the correlation matrix for all 24 items of the CBI-SS.It shows that many of the within-subscale correlations among the items exhibit coefficient values exceeding 0.5, indicating a high inter-item correlation within the subscales.However, it is noteworthy that the cross-subscale correlations are generally lower compared to the within-scale correlations.This observation lends support to the factorial structure of the CBI-SS.The only item diverging from this trend of generally high subscale inter-item correlation (0.384-0.394) is Item SRB 7, which demonstrates low correlations with all its low subscale items except for Item 8. Additionally, this item records one of the lowest correlations (0.212-0.215) with other subscale items.This poor performance is mirrored in its confirmatory factor analysis (CFA) results, where it simultaneously exhibits the lowest factor loading (0.540) and the highest residual variance (0.779).These figures suggest that the variance in the item may be inadequately explained by its latent factor.However, we did not remove the item from the model because the model fit statistics were satisfactory.Moreover, some authorities advise against removing items from models unless absolutely necessary, as even poorly fitting items can contribute to the fit of other, more suitable items 29 .

Criterion validity
The four factors demonstrate a negative correlation with both the General Academic Self-Efficacy scale and CGPA (Table 4).In addition, we examined the association between the CBI-SS subscales and self-reported burnout and perception of course as stressful using the Kruskal-Wallis test, a non-parametric method suitable for comparing medians across multiple groups.As Table 5 shows, students reporting definite burnout exhibited the highest median scores across all CBI-SS subscales (PB: 3.83, SRB: 3.67, FSRB: 3.17 range of responses within each category.A trend is observable where lower levels of self-reported burnout correspond with lower median scores in each subscale.This trend suggests a direct relationship between the degree of perceived burnout and the burnout subscales of the CBI-SS, for example, the category 'Not feeling burnout at all' consistently showed the lowest median scores for all subscales, which aligns with the expected outcomes of the CBI-SS in differentiating levels of burnout. The table also shows that students perceiving their course as 'stressful' recorded higher median scores on all subscales (PB: 3.67, SRB: 3.50, FSRB: 3.17, LRB: 3.17) compared to those not perceiving stress.The 'Maybe' response indicated uncertainty in course stress perception, yet this category also demonstrated elevated median scores, suggesting a potential link between uncertain stress perception and higher burnout scores.

Discussion
This study explored the psychometric properties of the student version of the Copenhagen Burnout Inventory among students of the Faculty of Arts at the Ekiti State University, Ado-Ekiti.Significantly, this study represents the first comprehensive assessment of the reliability and validity of the CBI-SS within a Nigerian academic context, and indeed within the African continent as a whole.Additionally, it extends the validation of the CBI-SS beyond its original validation study in Brazil.
The Cronbach's alpha coefficients of the CBI-SS were notably robust, ranging from 0.862 to 0.914 for all four subscales, with an overall value of 0.954 for the total scale.These reliability measures closely align with reported scores from the original Portuguese version of the CBI-SS (ranging from 0.875 to 0.931 for subscales and 0.957 for the total scale), as well as the Thai version (0.896 to 0.910 for all four subscales and 0.929 for the total scale) 11,16 .Moreover, both the Average Variance Extracted (AVE) and the Heterotrait-Monotrait (HTMT) ratios substantiate satisfactory levels of convergent and discriminant validity, respectively.
Significant and moderate to strong inter-factor correlations among the four CBI-SS dimensions indicate a considerable interdependence between the subscales.In light of this observation, prior scholars have proposed that an overall burnout score be derived from the four composite subscales of the CBI-SS 12,37 .Furthermore, an examination of the correlations between individual items in the four subscales revealed that they are moderately to strongly intercorrelated when compared to other items in the same subscale, but only weakly correlated when compared to items in other subscales.This observation provides significant support for the factorial design of the CBI-SS within our sample population.
The rationale behind Kristensen et al. 's development of the CBI as a replacement for the Maslach Burnout Inventory was, in part, that the MBI is a mismatch of three different components that should not be combined but studied in their own right 7 .They argued that depersonalization is a coping strategy for exhaustion while reduced personal accomplishment is a consequence of burnout.The high inter-subscale correlations and overall sound psychometric properties of the CBI-SS, as observed in our study, suggest it avoids this issue 11,37 .The fact that CBI-GS upholds the fundamental concept of burnout as energy depletion facilitates a clearer understanding of the causal relationship between burnout and other factors 7,37 .
Our study's results indicated that, similar to Campos et al. 's original adaptation of the CBI to a student population in Brazil, our data sample adequately fit the four-dimensional structure of the CBI-SS 11 .However, comparable adaptations in Portuguese and Brazilian contexts by Campos et al. and the Thai adaptation by Wongtrakul et al., achieving satisfactory fit necessitated the removal of one or more inventory items 11,16 .Specifically, in the Campos et al. adaptation, items 10 and 22 were excised due to low factor loading, resulting in only marginal fit improvements, as evidenced by the CFI and RMSEA values remaining below and above the respective thresholds.Similarly, the Thai adaptation involved the elimination of items 6, 10, and 17, along with the correlation of certain error terms during CFA.Our sample's fit surpassed these previous studies, maintaining adequate model fit parameters even with all items retained.However, the AVE for SRB and the factor loading for item 10 were below the accepted thresholds.When we removed item 10, the AVE for SRB increased above the 0.5 threshold.In the Thai version, items 4 and 7 were found to fit better in the SRB and Professional Burnout (PB) domains, respectively, rather than their original domains.This suggests that our model was more effective in fitting the sample than the two previous studies.We attribute this to two likely reasons.The first is that our study did not involve major translation, as Campos et al. provided an English version of their adaptation.The second reason could be our adaptation of the Likert-type options to ones more familiar to our students.Numerous authors have underscored the significance of language in the varying performances of psychometric measures across different countries and cultures [33][34][35] .Similar to the findings of Campos et al. and Wongkuratul et al., the factor loading for Item 10 was lower than the recommended threshold 15,16 .This phenomenon has also been observed in studies on the general CBI among Chinese workers 38 .Previous researchers posited that this low loading might have occurred because Item 10 was the only one phrased negatively, leading them to speculate that respondents might not have realized the scale was reversed 11,37 .While this is a plausible explanation, we propose that the wording of the item itself ('I have enough energy for family and friends during leisure time') is problematic.Firstly, this item presupposes that students allocate time for leisure and intend to spend it with friends and family.However, as Olubor et al. noted, 'The Nigerian students have not embraced a culture of leisure in their lifestyle.Their primary concern is excelling in their studies' 39 .Nigerian students' limited leisure time is often underutilized due to various factors, including lack of time, anxiety, academic workload, and religious activities 39,40 .Furthermore, the advent of the internet and social media means that most students would prefer spending their leisure time online rather than with friends and family.Additionally, many of our students live far from their families, and the concept of 'friends' may be ambiguous since, for many, their closest friends are fellow students.This ambiguity might contribute to confusion about how to respond to this item.Yeh et al., who researched the original CBI, observed similar uncertainties among Chinese workers, noting that the item loaded on a work-related factor for men and on a general burnout factor for women 37 .Therefore, we recommend the removal of this item in future applications of the Nigerian version of the CBI-SS.
The CBI-SS demonstrated good concurrent and discriminant validity, as evidenced by its performance against some other measures of academic performance and school stress.Both the GASE and the CGPA were negatively correlated with the CBI-SS subscales.Furthermore, the CBI-SS showed a significant association with self-reported burnout and perceived course stress.The significant relationships between CBI-SS subscale medians and selfreported burnout levels highlight the impact of burnout on various facets of student life.A clear association between the CBI-SS subscales and both the perceived course stress and self-reported burnout emphasizes the importance of addressing stress and burnout among students.These results contribute valuable insights to the understanding of burnout in academic contexts and underline the need for effective support and policy initiatives to support student well-being.

Conclusion
This study showcases the significance of the CBI-SS in evaluating student burnout within our student population.The study's findings demonstrate that the CBI-SS is a highly reliable and valid tool for evaluating student burnout, indicating its potential for effective assessment.The CBI-SS has shown strong Cronbach's alpha and McDonald's Omega (ω) coefficients and satisfactory levels of both convergent and discriminant validity in our sample.This indicates that it is a reliable and relevant tool for detecting burnout among students at Ekiti State University, Ado-Ekiti, Nigeria.The study's results emphasize the importance of considering cultural and contextual factors when using psychometric instruments, especially in terms of removing specific items to provide a good fit for the model.The strong correlation between the CBI-SS and measures of academic stress demonstrates the influence of burnout on students' lives.This study is a significant advancement in developing a more detailed and culturally aware understanding of academic burnout in the Nigerian student population.It also sets the stage for additional research and intervention in this essential element of student well-being and academic achievement.

Limitation
As an online study whose participants were recruited via WhatsApp, this study is subject to the well-known limitations inherent to such methodologies, including the potential for selection bias 41 .Specifically, students who are more active on social media platforms or possess internet-enabled mobile devices may have been disproportionately likely to participate, potentially skewing the study's representativeness of the broader student population.Additionally, the study's cross-sectional design presents another limitation, as it constrains the ability to determine causal relationships from the data.Finally, we lacked access to the university's general student data, which would have allowed us to assess how representative our cohort was of the entire student body.However, these limitations notwithstanding, the study boasts the distinct advantage of having a large sample size.This factor is particularly valuable, as it offers insights into a poorly studied phenomenon within an under-explored environment.

Table 1 .
Distribution of students by gender across various classes and courses in the faculty.*Record of gender distribution was not available.TMA Theatre and media arts.

Table 5 .
Association between the CBI-SS subscales and self-reported burnout and perception of course as stressful.