Health and Mental Well-Being of Academic Staff and Students in Thailand: Validation and Model Development

Abstract

A structural model of health and mental well-being among academic staff and students in Thailand was constructed and validated through confirmatory factor analysis (CFA). Data were obtained from 600 online questionnaires, equally distributed between staff (n = 300) and students (n = 300). Statistical analyses were undertaken in SPSS. Descriptive statistics were generated, internal reliability was assessed, and correlations were examined. The factor structure was first extracted through exploratory factor analysis (EFA). Model fit was subsequently assessed using CFA in LISREL. Five constructs were derived and validated: mental well-being (18 items), social participation (12 items), health literacy (28 items), work–life balance (10 items), and health behaviour (30 items). Convergent validity was demonstrated across all constructs. The final CFA model was found to exhibit a robust fit (χ² = 145.14, df = 62, p < 0.001, RMSEA = 0.047). Strong convergent validity and excellent fit indices were confirmed. Empirical evidence was therefore provided to support the model’s application in assessing health and mental well-being within Thai academic contexts.

1. Introduction

Health and mental well-being have been recognised as critical domains of academic inquiry due to their demonstrated influence on social sustainability, educational equity, and the resilience of human capital (Brown et al., 2023). The incorporation of mental health objectives within the United Nations Sustainable Development Goals (SDG 3: Good Health and Well-Being) has been acknowledged as reinforcing the imperative for evidence-based models that integrate psychological, physical, and social dimensions of health in educational contexts (Henderson & Loreau, 2023). Increasing global prevalence of stress, anxiety, depression, and burnout among university students and staff has been reported, with these conditions documented as having been exacerbated by digital learning transitions, economic constraints, and post-pandemic recovery demands (Lavrijsen et al., 2022).

Evidence from validated indicators of health and mental well-being (Duffy, 2023), including the Healthy Campus Framework in Canada and the Student Minds University Mental Health Charter in the United Kingdom, has been demonstrated to confirm that context-specific models can be applied to guide institutional policy (Brunner, 2023). International precedents have been recognised as demonstrating the potential for culturally adapted and empirically tested models to enhance well-being initiatives in Thailand’s higher education sector (Vien & Galik, 2024). The health and mental well-being of students and academic staff have been acknowledged as intrinsically associated with academic performance, institutional efficiency, and overall quality of life. The necessity of addressing both physical and psychosocial dimensions has been reinforced through the World Health Organization’s holistic definition of health and the OECD well-being indicators (Zhang et al., 2024). The importance of integrated approaches to educational well-being has been further underscored by the ASEAN Blueprint on Human Capital Development (Shang et al., 2024).

The complex realities of academic communities, in which stressors converge with socio-economic status, workload, institutional culture, and digital learning environments, have been overlooked in single-factor analyses (Meeks et al., 2023). The development of multivariable and validated models has been identified as providing a robust analytical framework for systematically mapping these interconnections. Through such mapping, the design of targeted and evidence-based strategies has been enabled for adoption by policymakers and university leaders to strengthen resilience and well-being outcomes (Patel et al., 2023). These models have also been acknowledged as informing interventions and establishing benchmarks for the continuous enhancement of higher education systems.

In Thailand, heightened national attention has been directed towards the health and mental well-being of academic staff and students, driven by the rising prevalence of stress, anxiety, depression, and burnout documented in surveillance data from the Ministry of Public Health and university health service records (Suwathanpornkul et al., 2023). National surveys have reported that more than one-third of university students experience moderate to severe psychological distress. Academic staff have been identified as encountering heavier workloads, intensified research performance expectations, and expanded administrative responsibilities resulting from higher education reforms (Liangruenrom et al., 2025).

Cultural dimensions, including collectivism, high power distance, and a face-saving ethos, have been identified as determinants of coping strategies and help-seeking behaviours (Jansem et al., 2025). These influences have been associated with the suppression of emotional expression and with a preference for informal support networks over professional services (Manochaiwuthikul et al., 2025). Stress exposure within Thai universities has been reported as being intensified by competitive academic environments, reinforced through performance-based funding mechanisms, and the pursuit of global rankings. Such pressures have been observed as constraining opportunities for restorative engagement (Thanoi et al., 2023).

Despite increasing awareness, substantial gaps have been reported in the detection, measurement, and management of mental health concerns within Thai higher education. Under-detection and under-reporting have been intensified by stigma and by cultural reluctance to disclose psychological distress, leading to a considerable proportion of cases remaining unaddressed (Ratanasiripong et al., 2024). An integrated and validated framework that simultaneously incorporates physical health, mental well-being, social participation, health literacy, work–life balance, and health behaviours among staff and students within a single model has not yet been developed (Phetphum et al., 2023).

In Thai academic contexts, social participation has been characterised as reflecting culturally specific norms of interaction, deference to authority, and collective involvement in faculty or student activities (Kantamaturapoj et al., 2023). These cultural dimensions have been reported as insufficiently represented in existing measurement instruments. Health literacy has been predominantly assessed using tools adapted from foreign contexts without full cultural calibration, which has limited their reliability within Thai populations. No robust, context-specific instrument has yet been developed for the evaluation of work–life balance or for the measurement of health behaviours in higher education (Schartner et al., 2023).

Despite increasing recognition of the importance of health and mental well-being in higher education, no validated and culturally appropriate tool or integrated model for assessing these dimensions among academic staff and students has yet been developed in Thailand (Ratanasiripong et al., 2024; Thanoi et al., 2023). This gap has been identified as critical, as existing instruments have frequently been adapted from foreign measures without full cultural calibration, thereby limiting their psychometric reliability in Thai contexts (McDonald & Nanni, 2023). The absence of such a model has been linked to ineffective interventions, inefficient resource allocation, and misalignment between institutional policies and the lived experiences of Thai academic communities. These constraints have been recognised as restricting the advancement of targeted, evidence-based strategies designed to strengthen both individual and institutional well-being (Aba Shaar et al., 2025).

The proposed framework has been constructed to integrate five interrelated variables—social participation, health literacy, work–life balance, health behaviours, and mental health—reflecting their collective influence on holistic well-being outcomes (Ratanasiripong et al., 2024; Manochaiwuthikul et al., 2025). The combined assessment of these dimensions has been acknowledged as enabling a more precise understanding of the intersections between personal, social, and institutional factors in shaping health trajectories within academic contexts (Thanapop et al., 2023). The validity of the scales has been established through CFA to ensure measurement accuracy. This procedure has been applied to capture construct relationships with both statistical and cultural precision prior to hypothesis testing.

The purpose of this study has been articulated as the development and validation of a culturally adapted model through CFA. Five interrelated determinants—social participation, health literacy, work–life balance, health behaviours, and mental health—have been integrated within the academic community of Thai higher education. Measurement instruments were designed to reflect Thai cultural and institutional contexts, and their psychometric properties were rigorously evaluated. High construct validity and internal consistency were demonstrated. Strong factor loadings across the five determinants were confirmed through CFA, indicating robust measurement precision. An empirically grounded model was finalised to capture the multidimensional nature of health and mental well-being. A validated framework has therefore been provided for the assessment and promotion of well-being among staff and students in higher education.

2. Materials and Methods

2.1. Data and Sample

A quantitative methodology was employed to examine the relationships among health literacy, work–life balance, and health behaviours among academic staff and university students in Thailand. Data were collected between September 2024 and February 2025 through electronic questionnaires and in-person administration across multiple university sites. Institutional diversity and geographical variation were ensured through the inclusion of public, private, and regional universities within the sampling frame. The target population was defined to include full-time undergraduate, graduate, and doctoral students, as well as academic staff comprising lecturers and support personnel affiliated with Thai universities. The minimum required sample size was calculated using Cochran’s formula for an unknown population, with a 95% confidence level and a 5% margin of error, resulting in a requirement of 600 respondents. To improve representativeness across university types and regions and to mitigate non-response bias, 700 questionnaires were distributed. A total of 600 fully completed and valid responses were obtained, yielding a response rate of 85.71%.

Participants were selected using a stratified random sampling technique, with stratification undertaken by university type (research-intensive versus teaching-focused). Proportional sampling was applied according to participant role (student versus academic staff). Eligibility criteria required a minimum enrolment or employment period of six months and comprehension of either Thai or English. Individuals on medical leave or those declining informed consent were excluded. The final sample consisted of 300 university students and 300 academic staff. The mean ages were 24.7 years for students and 38.3 years for academic personnel. Gender distribution was recorded as 61.4% female, 37.8% male, and 0.8% non-binary or preferring not to disclose. This sampling strategy has been supported in prior research, which has emphasised the importance of diverse representation across academic roles and university types when investigating lifestyle and mental health in Thai higher education.

2.2. Theoretical Scale Construction and Development

The theoretical construction of the scale was grounded in a multidimensional conceptualisation of health and mental well-being within the higher education context. Five core constructs—mental health, social participation, health literacy, work–life balance, and health behaviour—were identified for their interrelated influence on holistic student well-being (Alzadjali & Ahmad, 2024; Amoah et al., 2025; De et al., 2025; Garg et al., 2025). Social participation was theorised through social capital theory, in which engagement in formal and informal networks has been emphasised as a mechanism for fostering psychological resilience and a sense of belonging (Zarei et al., 2024). Health literacy was defined according to Petrič and Atanasova’s (2024) tripartite model of functional, interactive, and critical literacy and was regarded as enabling access to, comprehension of, and application of health-related information. Work–life balance was conceptualised within boundary theory, wherein equilibrium between academic, personal, and occupational demands has been viewed as necessary to reduce role strain and enhance well-being (Nitzsche et al., 2014). These constructs were incorporated as foundational dimensions of the theoretical scale and were treated as latent variables for empirical validation.

The theoretical model was designed to incorporate health behaviour, derived from the health belief model (Vasli et al., 2024), as an indicator of preventive and health-promoting actions, including physical activity, sleep hygiene, and nutrition, which have been recognised as contributing to both physical and psychological outcomes. Mental health was conceptualised as both a dependent construct and a central domain, operationalised through the dual-factor model, which accounts for the coexistence of well-being and the absence of distress (Agudelo-Hernández & Rojas-Andrade, 2024; Khan et al., 2024). These constructs were refined through iterative expert consultation, systematic synthesis of relevant literature, and alignment with validated instruments. Construct validity and content relevance for higher education populations were thereby ensured (Papavasileiou & Dimou, 2025; Stubin et al., 2024). The theoretical scale was constructed with emphasis on conceptual clarity, psychometric robustness, and contextual sensitivity. A model was developed to inform institutional interventions and policy recommendations intended to enhance student well-being outcomes. The procedures followed in the theoretical scale development are presented in

Table 1. Theoretical scale development procedures.

Stage	Recommended Procedure	Technique Implemented
1. Construct identification	Theoretical constructs were identified through an extensive review of relevant literature and the development of a coherent theoretical framework.	Foundational models, including the Health Belief Model, Social Capital Theory, and established frameworks of health literacy, were systematically reviewed.
2. Item generation	Items were developed to reflect and operationalise the dimensions of each construct.	Initial items were generated based on operational definitions, with between 10 and 15 items developed for each construct.
3. Content validity	Content relevance and clarity were assessed through expert evaluation.	The Index of Item–Objective Congruence (IOC) was applied, drawing on the evaluations of five subject-matter experts.
4. Item revision	Items identified as weak, based on expert feedback and IOC scores, were revised or eliminated.	Items receiving IOC scores below 0.50 were revised or removed.
5. Pilot testing	The instrument was pre-tested to evaluate clarity and reliability and to obtain user feedback.	A pilot study was conducted with 50 participants, comprising both staff and students, and qualitative feedback was obtained.
6. Questionnaire distribution	The full-scale survey was administered to the target population.	Online questionnaires were distributed to 300 staff members and 300 students across universities.
7. Construct validation	The factor structure and the interrelationships between constructs were assessed.	EFA and CFA were conducted.
8. Reliability testing	Internal consistency and measurement stability were evaluated.	Cronbach’s alpha, Composite Reliability (CR), and test–retest reliability were calculated.
9. Model development	The theoretical model was developed based on validated constructs.	A CFA structural model was constructed to estimate predictive pathways among the constructs.
10. Final refinement	The scale was refined to optimise its practical use and interpretability.	The item pool and scale instructions were finalised, and a user-friendly digital format was ensured.

.

2.3. Measures

Online questionnaires were employed to assess five core constructs: social participation, health literacy, work–life balance, health behaviour, and mental health. Each construct was measured at the ordinal level using a five-point Likert scale (1 = strongly disagree to 5 = strongly agree), thereby enabling standardised quantification of perceptions and behaviours. Social participation was operationalised through eight items adapted from established measures of social interaction and participation in health-related activities. Assessment was directed towards the frequency and quality of engagement in both formal and informal academic or community networks, with acceptable internal consistency demonstrated ($\alpha$ = 0.714).

Health literacy was assessed using ten items addressing access to health information and services, comprehension of health information and services, evaluation of health information and services, and application of health information and services. High internal consistency was demonstrated ($\alpha$ = 0.893). Work–life balance was measured using seven items, with assessment directed towards the perceived equilibrium between academic, personal, and occupational roles. Strong internal consistency was confirmed ($\alpha$ = 0.915). High reliability was therefore established for both constructs, indicating robust psychometric suitability within higher education contexts.

Health behaviour was assessed using nine items addressing exercise and physical activity, healthy eating, and safe sexual behaviour. Internal consistency was confirmed ($\alpha$ = 0.812). Mental health, defined as the dependent and central outcome variable, was measured using twelve items capturing mental state, mental competence, mental quality, and supporting factors, with high reliability demonstrated ($\alpha$ = 0.835). All measures were reviewed by experts through the Index of Item–Objective Congruence (IOC) prior to data collection and were subsequently validated using exploratory factor analysis and confirmatory factor analysis (CFA). Cronbach’s alpha scores exceeding 0.80 were recorded across all constructs, confirming internal reliability. This measurement framework was designed to generate actionable insights for educational policymakers and student support services, with the objective of improving health outcomes and strengthening institutional support systems in higher education.

2.4. Instruments

To measure the multidimensional constructs of health and mental well-being in higher education, a questionnaire instrument was employed following a rigorous process of item generation and validation. A comprehensive review of the literature was undertaken to identify the relevant domains—social participation, health literacy, work–life balance, health behaviour, and mental health. Items were then developed for each construct in alignment with these domains. Content validity was assessed through expert evaluation using the Index of Item–Objective Congruence (IOC). The IOC procedure was employed to evaluate rigorously the alignment between test items and their intended objectives. All items with scores below the 0.66 threshold were systematically excluded from the instrument (Turner & Carlson, 2003). A panel of five experts in health education, psychology, and psychometrics assessed each item for its alignment with the theoretical constructs. This process ensured strong theoretical fidelity and relevance to the target population, thereby providing a valid foundation for subsequent empirical testing.

The validated items were compiled into an online questionnaire using a secure digital platform and were administered to a stratified sample of 600 participants, comprising 300 university staff and 300 students from multiple higher education institutions. Online data collection was implemented to ensure broad geographical coverage and to enhance participant convenience, while anonymity and adherence to ethical standards were maintained in accordance with institutional review board approval. The questionnaire contained Likert-scale items for each construct, together with demographic variables to facilitate subgroup analysis. Inclusion of both staff and student populations was undertaken to strengthen the model’s generalisability across academic roles and to enable a more comprehensive understanding of well-being dynamics in higher education. The rigorous development and administration of the instrument were regarded as evidence of methodological integrity and as enhancing its potential contribution to educational and psychological research.

2.5. Proposed Model Modification

A health and mental well-being model (SP, HL, WLB, HB → MH) was developed in accordance with the methodological guidance of Asparouhov and Muthén (2009), Browne et al. (2018), and Chen et al. (2012). The model was constructed by integrating EFA and CFA within a structural framework. Cross-validation procedures were applied, together with assessments of reliability, convergent validity, discriminant validity, and multi-group invariance. Comparative analyses were undertaken between staff and student populations. Let the $p$ observed indicator e be represented in the vector $x \in {\mathbb{R}}^p$, and let $m$ common factors be denoted by the vector $f\in {\mathbb{R}}^m$. The model is then expressed in the following equations:

$$x=\mu + \mathsf{\Lambda }f+\epsilon ,f\sim \mathcal{N}(0,\mathsf{\Phi }),\epsilon \sim \mathcal{N}(0,\mathsf{\Psi })$$

(1)

with $\mathbb{E}[f\epsilon$^⊤] = 0.

Here, $\mathsf{\Lambda }(p\times m)$ denotes the factor loading matrix, $\mathsf{\Phi }(m\times m)$ represents the factor covariance matrix (oblique EFA), and ($\mathsf{\Psi })=$ diag$({\psi }_1$, …, ${\psi }_p)$ specifies the uniquenesses.

$$\mathsf{\Sigma }=\mathsf{\Lambda }\mathsf{\Phi }{\mathsf{\Lambda }}^{\mathbf{\top }}+\mathsf{\Psi }$$

(2)

The variance decomposition for indicator $i$ is given as follows:

$$\mathrm{V}\mathrm{a}\mathrm{r}(x_i={ h}_i^2+ {\psi }_1, { h}_i^2=\sum _{j=1}^m{\lambda }_{ij}^2) \mathrm{i}\mathrm{f} \mathsf{\Phi }=\mathbf{I} ; \mathrm{o}\mathrm{t}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{w}\mathrm{i}\mathrm{s}\mathrm{e}{ h}_i^2=(\mathsf{\Lambda }\mathsf{\Phi }{\mathsf{\Lambda }}^{\mathbf{\top }}{)}_{ii}$$

(3)

An orthogonal or oblique rotation $\mathbf{T}$ is applied such that the rotated loading matrix is expressed as $\mathsf{\Lambda }= \mathsf{\Lambda }\mathbf{T}$, yielding a simpler structure. In the oblique case, the rotated factor covariance matrix is given by:

$$\tilde{\mathsf{\Phi }}={[\mathbf{T}}^{-1}\mathsf{\Phi }{\mathbf{T}}^{-1}].$$

(4)

Let $k_c$ denote the number of observed indicators for each construct $c\in =\left\{\mathrm{S}\mathrm{P},\mathrm{H}\mathrm{L},\mathrm{W}\mathrm{L}\mathrm{B},\mathrm{H}\mathrm{B},\mathrm{M}\mathrm{H}\right\}$. Accordingly, the measurement specification can be expressed as follows:

$$\begin{gathered} \mathrm{S}\mathrm{o}\mathrm{c}\mathrm{i}\mathrm{a}\mathrm{l} \mathrm{s}\mathrm{u}\mathrm{p}\mathrm{p}\mathrm{o}\mathrm{r}\mathrm{t} \left(SP\right):S{P}_1,\dots ,S{P}_{kSP}; \\ \mathrm{H}\mathrm{e}\mathrm{a}\mathrm{l}\mathrm{t}\mathrm{h} \mathrm{l}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{c}\mathrm{y} \left(HL\right); H{L}_1,\dots ,H{L}_{kHL}; \\ \mathrm{W}\mathrm{o}\mathrm{r}\mathrm{k}–\mathrm{l}\mathrm{i}\mathrm{f}\mathrm{e} \mathrm{b}\mathrm{a}\mathrm{l}\mathrm{a}\mathrm{n}\mathrm{c}\mathrm{e} \left(WLB\right):WK{B}_1,\dots ,WL{B}_{kWLB}; \\ \mathrm{H}\mathrm{e}\mathrm{a}\mathrm{l}\mathrm{t}\mathrm{h} \mathrm{b}\mathrm{e}\mathrm{h}\mathrm{a}\mathrm{v}\mathrm{i}\mathrm{o}\mathrm{u}\mathrm{r}\mathrm{s} \left(HB\right):H{B}_1,\dots ,H{B}_{kHB}; \\ \mathrm{M}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{l} \mathrm{h}\mathrm{e}\mathrm{a}\mathrm{l}\mathrm{t}\mathrm{h} \left(MH\right): M{H}_1,\dots ,M{H}_{kMH}. \end{gathered}$$

(5)

For indicator $y_{c\mathcal{l}}$ ($\mathcal{l}=1,\dots ,k_c)$:

$$y_{c\mathcal{l}}= {\tau }_{c\mathcal{l}}+ {\lambda }_{c\mathcal{l}}{\eta }_c+{ϵ}_{c\mathcal{l}}, {ϵ}_{c\mathcal{l}} \sim \mathcal{N}\left(0,{\theta }_{c\mathcal{l}}\right),$$

(6)

with ${\eta }_c$ the latent score for construct $c$. Stacked (all indicator $\mathit{y}$):

$$\mathit{y}=\mathsf{\lambda }+\mathsf{\Lambda }\eta +\mathit{ϵ}, \mathit{ϵ}\sim \mathcal{N}\left(0,\mathsf{\Theta }\right),\mathsf{\Theta }=\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{g}\left({\theta }_i\right).$$

(7)

For model identification, the first loading of each construct is fixed to unity, such that 1: ${\lambda }_{c1}=1$. The variance of each latent construct ${\eta }_c$ is then estimated as ${\varphi }_{cc}$.

Let groups be indexed by $g \in$ {staff, students}. For groups $g$, the measurement and structural parameters are estimated separately, enabling the assessment of measurement invariance and structural equivalence across groups. The measurement and structural model for group $g$ is expressed as:

$${\mathit{y}}^{\left(g\right)}={\mathsf{\tau }}^{\left(g\right)}+{\mathsf{\Lambda }}^{\left(g\right)}{\eta }^{\left(g\right)}+ {\mathit{ϵ}}^{\left(g\right)}, {\mathit{ϵ}}^{\left(g\right)}\sim \mathcal{N}\left(0,{\Theta }^{\left(g\right)}\right).$$

(8)

$${\mathit{\eta }}^{\left(g\right)}={\mathit{\beta }}^{\left(g\right)}+{\mathit{\xi }}^{\left(g\right)}{\mathit{\zeta }}^{\left(g\right)}+\mathrm{C}\mathrm{o}\mathrm{v}\left({\mathit{\xi }}^{\left(g\right)}\right)={\mathsf{\Phi }}_{\xi }^{\left(g\right)}, \mathrm{V}\mathrm{a}\mathrm{r}\left({\mathit{\zeta }}^{\left(\mathit{g}\right)}\right)={\psi }_{\zeta }^{\left(g\right)}$$

(9)

In the configural model, ${\mathsf{\Lambda }}^{\left(\mathit{g}\right)}$ shares an identical pattern of fixed and free elements across groups, with all parameters freely estimated. In the metric invariance model, the factor loadings are constrained to equality across groups, such that ${\mathsf{\Lambda }}^{\left(\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{f}\mathrm{f}\right)}={\mathsf{\Lambda }}^{\left(\mathrm{s}\mathrm{t}\mathrm{u}\mathrm{d}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{s}\right)}$. Scalar invariance is imposed by additionally constraining the intercept, ${\mathsf{\tau }}^{\left(\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{f}\mathrm{f}\right)}={\mathsf{\tau }}^{\left(\mathrm{s}\mathrm{t}\mathrm{u}\mathrm{d}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{s}\right)}$. Residual invariance is then tested by equating the error variances, ${\mathsf{\Theta }}^{\left(\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{f}\mathrm{f}\right)}={\mathsf{\Theta }}^{\left(\mathrm{s}\mathrm{t}\mathrm{u}\mathrm{d}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{s}\right)}$. For structural comparisons, the equality of regression paths is examined by testing ${\mathit{\beta }}^{\left(staff\right)}={\mathit{\beta }}^{\left(students\right)}$. Optionally, equivalence constraints may also be placed on the latent covariances and disturbance variances, such that ${\mathsf{\Phi }}_{\xi }^{\left(g\right)}$, ${\psi }_{\zeta }^{\left(g\right)}$ are set equal across groups.

As a useful derived quantity for reporting, the composite reliability (CR) was developed for each construct $c$. It is defined as:

$${\mathrm{C}\mathrm{R}}_c={\displaystyle \frac{({\sum }_{\mathcal{l}}{\lambda }_{c\mathcal{l}})}{({\sum }_{\mathcal{l}}{\lambda }_{c\mathcal{l}}{)}^{2 }+ {\sum }_{\mathcal{l}}{\theta }_{c\mathcal{l}}}}$$

(10)

The average variance extracted (AVE) for construct $c$ is defined as:

$$\mathrm{A}\mathrm{V}\mathrm{E}={\displaystyle \frac{({\sum }_{\mathcal{l}}{\lambda }_{c\mathcal{l}}^2)}{({\sum }_{\mathcal{l}}{\lambda }_{c\mathcal{l}}^2{)}^{2 }+ {\sum }_{\mathcal{l}}{\theta }_{c\mathcal{l}}}}$$

(11)

Discriminant validity was assessed using the Fornell–Larcker criterion. For each construct $c$, the square root of the average variance extracted, $\sqrt{{AVE}_c, }$ must exceed the absolute value of its correlations with all other constructs $\acute{c}\ne c$:

$$\sqrt{{AVE}_c}> \mid {\mathsf{\varphi }}_{c\acute{c}}\mid \mathrm{for} \mathrm{all} \acute{c}\ne c$$

(12)

where ${\mathsf{\varphi }}_{c\acute{c}}$ denotes the estimated correlation between constructs $\acute{c}\ne c.$ This condition indicates that a construct shares greater variance with its own indicators than with other latent constructs, thereby supporting discriminant validity.

2.6. Analyses

Data analysis was conducted using a combination of descriptive and inferential statistical techniques to ensure rigorous validation of the developed measurement model. All responses were coded and processed in SPSS (Version 28) for data cleaning, treatment of missing values, testing of normality, and computation of descriptive statistics, including means, standard deviations, skewness, and kurtosis. Internal consistency reliability was assessed using Cronbach’s alpha, with all constructs recording values above the accepted threshold of 0.70. The underlying factor structure was examined through EFA using principal axis factoring with promax rotation, which was considered appropriate for correlated constructs. Sampling adequacy was confirmed by the KMO value exceeding 0.80 and by Bartlett’s test of sphericity at p < 0.001. EFA results were applied to guide item reduction, with items demonstrating factor loadings below 0.50 or cross-loadings being excluded, thereby refining the factor structure in preparation for CFA.

CFA was conducted in LISREL 10.3 to validate the factor structure identified during the EFA and to evaluate the measurement model’s overall goodness of fit. Relationships between observed variables and latent constructs were examined, and construct validity was assessed through model fit indices, including χ²/df, CFI > 0.90, RMSEA < 0.08, and SRMR < 0.08 (Hu & Bentler, 1999). Convergent and discriminant validity were evaluated using AVE > 0.50) and CR > 0.70 (Goodwin, 1999; Mueller et al., 1998). Model re-specification was undertaken where modification indices provided theoretically justifiable adjustments. This dual-phase analytical process was applied to ensure a statistically robust and theoretically coherent validation of the scale, thereby contributing to the development of an empirically grounded model of health and mental well-being in higher education.

3. Results

3.1. Sample Characteristics

The total sample comprised 600 participants, divided equally between university students (n = 300) and academic staff (n = 300). Among the staff, 85 individuals (28.33%) were employed in academic roles, while 215 (71.67%) held operational positions. Within the student cohort, 90.67% were enrolled in undergraduate programmes, 4.67% in master’s programmes, and 4.67% in doctoral programmes. Disciplinary distribution across three academic clusters—social sciences and humanities, health sciences, and science and technology—was observed to differ significantly (χ² = 6.829, p = 0.033), indicating potential variation in mental health and well-being across fields of study or employment.

The gender distribution was recorded as 21.83% male and 67.83% female, with 8.17% identifying as LGBTQ+ and 2.17% choosing not to disclose. No statistically significant gender differences were observed between students and staff (χ² = 4.584, p = 0.205). As expected, age distribution differed markedly: the majority of students were under 25 years of age (89.33%), whereas staff were concentrated in older age brackets, particularly 35–44 years (38.33%) and 45–54 years (29.33%). Financial well-being also varied significantly, with students more likely to report sufficiency or savings (51.33%) compared with staff (34%), while staff reported higher debt levels, including 25.67% indicating financial insufficiency and indebtedness (χ² = 59.000, p < 0.001). Living arrangements differed significantly as well, with staff predominantly residing with family (78%), while students reported more varied arrangements, including living alone (23.67%) or with friends (14.67%) (χ² = 39.637, p < 0.001). The composition of student and staff samples is summarised in

Table 2. Sample characteristics.

General Information	Student		Staff		Total		χ2	p-Value
1. Sample characteristics
Academic staff	–	–	85	28.33	85	14.17	–	–
Support staff	–	–	215	71.67	215	35.83
Undergraduate student	272	90.67	–	–	272	45.33
Graduate student	14	4.67	–	–	14	2.33
PhD student	14	4.67	–	–	14	2.33
2. Which field does your current faculty or field of study belong to?
Humanities and Social Sciences	149	49.67	130	43.33	279	46.50	6.829	0.033 *
Health Science	100	33.33	93	31.00	193	32.17
Science and Technology	51	17.00	77	25.67	128	21.33
3. Gender
Male	65	21.67	66	22.00	131	21.83	4.584	0.205
Female	198	66.00	209	69.67	407	67.83
LGBTQ+	27	9.00	22	7.33	49	8.17
Not Specified	10	3.33	3	1.00	13	2.17
4. Level of Age
Under 25 years old	268	89.33	6	2.00	274	45.67	–	–
25–34 years old	16	5.33	74	24.67	90	15.00
35–44 years old	10	3.33	115	38.33	125	20.83
45–54 years old	6	2.00	88	29.33	94	15.67
55 years old and over	–	–	17	5.67	17	2.83
4. Current overall income and expenditure
Sufficient/with savings	154	51.33	102	34.00	256	42.67	59.000	0.000 *
Sufficient/no savings	111	37.00	96	32.00	207	34.50
Not sufficient/without debt	23	7.67	25	8.33	48	8.00
Insufficient/in debt	12	4.00	77	25.67	89	14.83
5. With whom do you reside for most of the time?
Family	185	61.67	234	78.00	419	69.83	39.637	0.000 *
Friend	44	14.67	4	1.33	48	8.00
Live alone	71	23.67	62	20.67	133	22.17

Note: * = p < 0.05.

.

Moderate to strong positive relationships among the key constructs—social participation, health literacy, work–life balance, health behaviour, and mental health—were identified through the intercorrelation matrix, thereby indicating convergent validity and internal coherence within the theoretical model. Most variables demonstrated moderate to strong positive associations, particularly between knowledge, checking, and application ($r$ > 0.69), suggesting close interrelation among these factors. Dimensions of mental health—including mind, ability, quality, and support—were also found to be strongly interconnected ($r$ ≈ 0.55–0.66), indicating a coherent cognitive–supportive domain. By contrast, lifestyle factors such as activity, food, and sex exhibited weaker correlations with cognitive and skill-based variables, suggesting greater independence of these dimensions within the dataset. The intercorrelations among variables for both academic staff and students are summarised in

Table 3. Intercorrelations among variables.

Variable	1	2	3	4	5	6	7	8	9	10	11	12	13	14
1. Mental state	1.000
2. Mental capacity	0.613 *	1.000
3. Mental quality	0.472 *	0.657 *	1.000
4. Supporting factors	0.550 *	0.631 *	0.568 *	1.000
5. Social interaction	0.529 *	0.700 *	0.549 *	0.633 *	1.000
6. Participation in health activities	0.454 *	0.506 *	0.325 *	0.538 *	0.585 *	1.000
7. Work–life balance	0.514 *	0.495 *	0.447 *	0.558 *	0.530 *	0.448 *	1.000
8. Access to health information and services	0.338 *	0.438 *	0.388 *	0.477 *	0.562 *	0.356 *	0.399 *	1.000
9. Understanding of health information and services	0.361 *	0.513 *	0.446 *	0.480 *	0.610 *	0.333 *	0.416 *	0.717 *	1.000
10. Assessment of health information and services	0.346 *	0.522 *	0.448 *	0.473 *	0.605 *	0.327 *	0.436 *	0.681 *	0.754 *	1.000
11. Application of health information and services	0.440 *	0.518 *	0.443 *	0.513 *	0.641 *	0.414 *	0.512 *	0.674 *	0.694 *	0.792 *	1.000
12. Exercise and physical activity	0.294 *	0.259 *	0.201 *	0.254 *	0.285 *	0.332 *	0.264 *	0.221 *	0.191 *	0.254 *	0.291 *	1.000
13. Healthy eating	0.309 *	0.304 *	0.308 *	0.350 *	0.365 *	0.290 *	0.361 *	0.298 *	0.287 *	0.338 *	0.374 *	0.505 *	1.000
14. Safe sexual behaviour	0.173 *	0.230 *	0.255 *	0.261 *	0.298 *	0.172 *	0.346 *	0.269 *	0.298 *	0.339 *	0.316 *	0.039	0.229 *	1.000
$\overline{X}$	3.70	4.05	4.40	4.11	4.13	3.44	3.93	4.14	4.20	4.24	4.11	2.33	3.04	4.43
SD	0.69	0.61	0.58	0.67	0.63	0.99	0.71	0.63	0.60	0.62	0.62	0.75	0.74	0.60

Note: * = p < 0.05.

.

3.2. Initial Model Development

Data readiness was established through the treatment of missing values using FIML. Univariate outliers were screened using the criterion ∣$z$∣ > 3.29, while multivariate outliers were identified through Mahalanobis distance with a significance threshold of $p$ < 0.001. Normality was examined by evaluating skewness and kurtosis, and robust MLR was applied where assumptions were violated. The dataset was partitioned for model development and validation, either through a 70/30 split or by employing five-fold cross-validation. Sampling adequacy was confirmed by a KMO value of at least 0.80, a statistically significant Bartlett’s test of sphericity (p < 0.001), and adherence to the criterion of a minimum sample size equivalent to at least ten times the number of observed indicators (n ≥ 600, where feasible).

EFA was conducted on the development set. The factor structure was identified using parallel analysis and the MAP test. Factors were extracted through maximum likelihood estimation, and oblique rotation was applied as the preferred method. Items were retained if they demonstrated a primary loading of at least 0.50, a cross-loading of no more than 0.30, and a communality ($h$²) of at least 0.40. Refinement of the item pool was additionally guided by conceptual coherence across the constructs of SP, HL, WLB, HB, and MH. All item exclusions were systematically documented and explicitly justified.

The initial CFA was conducted on the development set. Model identification was achieved either by constraining the first loading of each construct to unity or by fixing the latent variance to unity. Estimation was performed using robust MLR, with FIML applied to address missing data. Model fit was evaluated against established benchmarks: CFI and TLI values of at least 0.95 (with 0.90 considered acceptable), RMSEA of 0.06 or lower (with 0.08 considered acceptable) including the 90% confidence interval, and SRMR of 0.08 or lower. Modification indices and correlated residuals were incorporated only where theoretically justified to minimise the risk of post hoc overfitting.

Reliability and convergent validity were assessed using standardised loadings (λ ≥ 0.60; ≥0.50 acceptable), composite reliability (CR ≥ 0.70), coefficient omega (≥0.70), and average variance extracted (AVE ≥ 0.50) for each construct. Discriminant validity was established through the Fornell–Larcker criterion, whereby the square root of each construct’s AVE exceeded its inter-construct correlations, and through the HTMT, which was required to remain below 0.85 under the strict criterion and below 0.90 under the lenient criterion.

3.3. Measurement Model

The analysis of the measurement model across the five core constructs was found to demonstrate overall adequacy for CFA; however, several indicators exhibited non-normality, as evidenced by skewness, kurtosis, and Shapiro–Wilk p-values. Variables relating particularly to health literacy (understanding, evaluation, and application of health information) and to mental health quality were characterised by moderate to high negative skewness and leptokurtic distributions, with significant departures from normality in the total sample (p < 0.05). For example, the quality of mental health and safe sexual behaviour recorded kurtosis values exceeding –2.8, with statistically significant p-values (p = 0.000 and p = 0.002, respectively), indicating peaked and left-skewed distributions.

Despite these deviations, non-normality has been recognised as common in psychological data (Byrne, 2013), and estimation methods such as robust maximum likelihood or asymptotic covariance matrices in CFA (LISREL) were considered appropriate for addressing these violations. By contrast, variables such as physical activity and healthy eating demonstrated near-normal distributions (p > 0.80), indicating more balanced response patterns. The measurement model results of health and mental well-being components for academic staff and students are presented in

Table 4. Measurement model results.

Variable	Student Groups			Staff Groups			Total Groups
	Skewness	Kurtosis	p-Value	Skewness	Kurtosis	p-Value	Skewness	Kurtosis	p-Value
1. Mental state	−0.463	−0.755	0.676	−0.394	−0.714	0.717	−0.640	−1.041	0.474
2. Mental capacity	−1.118	−1.727	0.121	−0.752	−1.210	0.362	−1.337	−2.084	0.052
3. Mental quality	−2.743	−2.940	0.000	−2.728	−2.986	0.000	−2.893	−2.995	0.000
4. Supporting factors	−1.230	−1.866	0.080	−1.136	−1.645	0.136	−1.677	−2.534	0.030
5. Social interaction	−1.549	−2.101	0.053	−1.110	−1.697	0.128	−1.881	−2.738	0.050
6. Participation in health activities	−0.905	−1.709	0.154	−0.650	−1.244	0.374	−1.089	−2.203	0.056
7. Work–life balance	−0.760	−1.231	0.351	−0.626	−1.046	0.476	−0.983	−1.685	0.149
8. Access to health information and services	−1.473	−2.185	0.051	−1.165	−1.730	0.114	−1.859	−2.816	0.008
9. Understanding of health information and services	−1.547	−2.202	0.027	−1.364	−2.070	0.050	−2.056	−2.602	0.018
10. Assessment of health information and services	−1.560	−2.417	0.016	−1.662	−2.302	0.018	−2.282	−2.763	0.006
11. Application of health information and services	−1.631	−2.305	0.029	−1.092	−1.523	0.173	−1.914	−2.795	0.032
12. Exercise and physical activity	0.091	−0.203	0.976	0.150	−0.494	0.875	0.172	−0.620	0.813
13. Healthy eating	−0.016	−0.054	0.998	−0.040	−0.019	0.999	−0.061	−0.120	0.991
14. Safe sexual behaviour	−1.615	−2.280	0.020	−1.685	−2.301	0.017	−2.327	−2.864	0.002

.

The development and validation of the mental well-being scale were undertaken, resulting in the identification of four distinct components: (1) mental state, (2) mental competence, (3) mental quality, and (4) supporting factors. The finalised scale consisted of 18 items. Confirmatory factor analysis demonstrated that the proposed measurement model was consistent with the empirical data (χ² = 270.53, df = 116, p < 0.001, χ²/df = 2.332, RMSEA = 0.047, SRMR = 0.034, GFI = 0.95, CFI = 0.99, NFI = 0.98), thereby providing evidence of construct validity. Analysis of the standardised factor loadings indicated that all items satisfied established psychometric thresholds, with loadings ranging from 0.64 to 0.78.

The development and validation of the social interaction scale were undertaken, resulting in the identification of two components: (1) social interaction and (2) participation in health-related activities. The finalised scale comprised 12 items. Confirmatory factor analysis demonstrated that the proposed measurement model was consistent with the empirical data (χ² = 59.62, df = 24, p < 0.001, χ²/df = 2.484, RMSEA = 0.050, SRMR = 0.027, GFI = 0.98, CFI = 1.00, NFI = 0.99), thereby providing evidence of construct validity. Examination of the standardised factor loadings indicated that all items satisfied established psychometric thresholds, with loadings ranging from 0.57 to 0.99.

The development and validation of the health literacy scale were undertaken, resulting in the identification of four components: (1) access to health information and services, (2) understanding of health information and services, (3) evaluation of health information and services, and (4) application of health information and services. The finalised scale comprised 28 items. Confirmatory factor analysis demonstrated that the proposed measurement model was consistent with the empirical data (χ² = 815.38, df = 328, p < 0.001, χ²/df = 2.486, RMSEA = 0.050, SRMR = 0.035, GFI = 0.91, CFI = 0.99, NFI = 0.99), thereby providing robust evidence of construct validity. Examination of the standardised factor loadings indicated that all items satisfied established psychometric thresholds, with values ranging from 0.64 to 0.83.

The development and validation of the work–life balance scale were undertaken, resulting in a final instrument comprising 10 items. Confirmatory factor analysis demonstrated that the proposed measurement model was consistent with the empirical data (χ² = 29.66, df = 22, p = 0.127, χ²/df = 1.348, RMSEA = 0.024, SRMR = 0.016, GFI = 0.99, CFI = 1.00, NFI = 1.00), thereby providing evidence of construct validity. Examination of the standardised factor loadings indicated that all items satisfied established psychometric thresholds, with values ranging from 0.57 to 0.80.

The development and validation of the health behaviour scale were undertaken, resulting in the identification of three components: (1) exercise and physical activity, (2) healthy eating, and (3) safe sexual behaviour. The finalised scale comprised 30 items. Confirmatory factor analysis demonstrated that the proposed measurement model was consistent with the empirical data (χ² = 1006.07, df = 382, p < 0.001, χ²/df = 2.634, RMSEA = 0.052, SRMR = 0.057, GFI = 0.92, CFI = 0.97, NFI = 0.96), hereby providing evidence of construct validity. Examination of the standardised factor loadings indicated that all items satisfied established psychometric thresholds, with values ranging from 0.30 to 0.76.

3.4. CFA Structural Model

A total of 40% of the variance in work–life balance, 84% of the variance in health behaviour, and 76% of the variance in mental health were explained by the model, with the strongest predictive power observed for health behaviour. Mental health was found to be most strongly influenced by social participation (total effect = 0.83), which also exerted a substantial direct effect on work–life balance. Health behaviour was most strongly predicted by health literacy (total effect = 0.72). Direct positive effects of work–life balance were identified for both health behaviour (0.39) and mental health (0.22), underscoring its role as a key determinant. The validity of the structural model, as confirmed through confirmatory factor analysis, is presented in

Table 5. CFA structural model.

Variable	Work–Life Balance $({\mathbf{R}}^2$ = 0.40)			Health Behaviour $({\mathbf{R}}^2$ = 0.84)			Mental Health $({\mathbf{R}}^2$ = 0.76)
	DE	IE	TE	DE	IE	TE	DE	IE	TE
Social participation	0.49 *	–	0.49 *	–	0.19 *	0.19 *	0.72 *	0.11 *	0.83 *
Health literacy	0.17 *	–	0.17 *	0.65 *	0.07 *	0.72 *	–	0.04 *	0.04 *
Work–life balance	–	–	–	0.39 *	–	0.39 *	0.22 *	–	0.22 *

Note: * = p < 0.05.

.

A good fit of the CFA structural model was demonstrated (χ² = 145.14, df = 62, p < 0.001, RMSEA = 0.047), thereby supporting the construct validity of the latent variables. Health literacy and social participation were shown to be well measured by their respective indicators, with strong factor loadings (>0.77 for health literacy and >0.66 for social participation), and both exerted significant structural paths towards health behaviour, work–life balance, and mental health. The model confirmed that social participation exerted the strongest total effect on mental health, that health literacy most strongly predicted health behaviour, and that work–life balance served as a mediating link between the predictors and mental health outcomes. The validation of the CFA model is illustrated in Figure 1.

3.5. Hypothesis Testing

A good fit of the hypothesised relationship model was demonstrated at each stage of invariance testing, with RMSEA consistently reported at 0.060, GFI at 0.93, and χ²/df values maintained at approximately 2, all of which were within acceptable thresholds. In Step 1 (configural invariance), the basic factor structure was confirmed as consistent across groups. In Step 2 (metric invariance), no significant difference from Step 1 was detected (Δχ² = 17.73, p = 0.059), indicating equivalence of factor loadings across groups. In Step 3 (structural invariance), no significant difference from Step 2 was observed (Δχ² = 11.26, p = 0.081), thereby confirming the invariance of the structural paths. Collectively, these findings indicated that the measurement and structural relationships within the model were stable and comparable, providing strong evidence of validity across groups. The results of the hypothesis testing are presented in

Table 6. Hypothesis testing.

Hypothesised Relationship Model	Fit Index						Invariance Condition
	χ2	df	p-Value	RMSEA	χ2/df	GFI	Δχ2	Δdf	p-Value
Step 1: Configural invariance	315.09	151	0.00	0.060	2.073	0.93	–	–	–
Step 2: Metric invariance	332.82	161	0.00	0.060	2.067	0.93	17.73 (2→1)	10 (2→1)	0.059
Step 3: Structural invariance	344.08	167	0.00	0.060	2.060	0.93	11.26 (3→2)	6 (3→2)	0.081

Note: → model path.

.

A statistically significant mean difference between groups was identified only for social participation (mean difference = 0.158, t = 2.694, p = 0.007), indicating meaningful variation in this construct. No significant mean differences were detected for the other latent variables—health literacy, work–life balance, health behaviour, and mental health—with all p-values exceeding 0.05. These findings suggest that, apart from social participation, the groups were comparable in their levels of the remaining constructs. The disparity observed in social participation may be attributable to contextual or cultural factors specific to each group that influence engagement in social activities. Overall, stability across most constructs was confirmed, with social participation identified as the sole area of notable difference. The mean differences of the latent variables are presented in

Table 7. The mean difference of latent variables.

Latent Variable	Mean Difference	$\mathit{\delta }$	t-Value	p-Value
Social participation	0.158	0.059	2.694	0.007 *
Health literacy	0.003	0.045	0.074	0.941
Work–life balance	0.074	0.059	1.276	0.203
Health behaviour	0.062	0.041	1.516	0.130
Mental health	0.025	0.044	0.572	0.567

Note: * = p < 0.05.

.

4. Discussion

4.1. Practical Implications

The validated model, confirmed through EFA and CFA, was found to demonstrate strong construct validity and internal consistency across all five dimensions, thereby providing robust psychometric support for the assessment of health and mental well-being in higher education. Statistically significant differences between academic staff and students were observed, particularly in work–life balance and mental health. Lower work–life balance scores were recorded among staff, attributed to increased administrative and personal responsibilities, whereas higher levels of mental health concerns were reported among students, associated with reduced social participation and lower health literacy.

The CFA structural model fit indices (RMSEA = 0.047, CFI > 0.90) confirmed the structural integrity of the model across both groups. Moderate correlations between health behaviour and mental health outcomes were identified in both populations, underscoring the significance of health behaviour as a modifiable determinant (Hu & Bentler, 1999). These findings highlight the relevance of the model for both theoretical validation and practical application in institutional well-being monitoring and policy formulation (Haybron & Tiberius, 2015; Wampler & Touchton, 2019; Cook & Davíðsdóttir, 2021).

From a practical perspective, the validated tool has been positioned as an evidence-based framework through which universities may assess and address the distinct well-being needs of staff and students. The significant roles of social participation and health literacy in predicting mental health were emphasised, thereby underscoring the value of co-curricular programmes, peer networks, and targeted health education (Henderson & Loreau, 2023). For students, psychological resilience was shown to be strengthened through structured mentorship schemes, orientation workshops embedding health literacy, and inclusive peer-support clubs that promote belonging.

For staff, occupational stress was mitigated through flexible scheduling, wellness initiatives, and institutional strategies aimed at restoring work–life integration (Yadav et al., 2022). Concrete practices such as faculty–student engagement programmes, digital well-being platforms, and community-based outreach were recommended to translate the findings into actionable policies. The strong internal consistencies ($\alpha$ > 0.80) and the demonstrated generalisability of the model across roles reinforced its value for longitudinal monitoring, early risk detection, and personalised interventions. The model was positioned as both a diagnostic and strategic resource for higher education leaders, thereby enabling the development of health-promoting campuses that support academic productivity and institutional sustainability.

4.2. Theoretical Contributions

A significant theoretical contribution was achieved through the empirical validation of a multidimensional model of health and mental well-being in higher education using confirmatory factor analysis (CFA). The five latent constructs—social participation, health literacy, work–life balance, health behaviour, and mental health—were found to be psychometrically robust and conceptually interrelated across both academic staff and student populations. CFA results confirmed that the measurement model achieved strong goodness-of-fit indices (CFI > 0.90, RMSEA < 0.08), thereby affirming the theoretical coherence of the structure (Hu & Bentler, 1999). The model was further reinforced through the integration of social capital theory (Putnam et al., 2004), health literacy theory (Nutbeam, 2000), work–family theory (Clark, 2000), and the health belief model (Rosenstock, 1974) within a unified, education-specific framework. This integration bridged previously disconnected domains of health behaviour and institutional engagement, resulting in a comprehensive conceptualisation of well-being in academic environments.

Theoretical variations between students and academic staff were identified by the model, with differential structural pathways observed. For staff, work–life balance was shown to be a stronger predictor of mental health, reflecting the dual responsibilities embedded within professional and personal domains. For students, social participation and health literacy exerted greater influence, highlighting the importance of peer connection and health knowledge in shaping well-being. These findings contributed to the literature on context-sensitive health psychology by validating the ways in which occupational and developmental stages influence the salience of well-being determinants (Keyes, 2002). The study was therefore found not only to validate a psychometric model but also to extend theoretical understanding of how distinct academic roles affect health outcomes, thereby supporting the development of tailored interventions in higher education policy and student affairs. This dual-level validation was shown to enhance the theoretical generalisability of the model while recognising group-specific dynamics.

4.3. Future Implications and Policy Initiatives

The validated model presented in this study was found to highlight the urgent need for integrated well-being policies within higher education institutions. These policies were framed to extend beyond the treatment of mental health concerns and to encompass preventive, structural, and educational dimensions. Comprehensive strategies were recommended to promote social participation. Inclusive clubs, peer support networks, and faculty–student engagement programmes were proposed to foster belonging. Such initiatives were identified as critical for students transitioning into university life and for academic staff balancing teaching and research responsibilities. Health literacy was recommended for integration within academic curricula and orientation programmes, thereby equipping staff and students with the knowledge and confidence required to access, evaluate, and apply health-related information effectively (Nutbeam, 2000). Equitable access to mental health services was emphasised, with digital platforms, crisis intervention resources, and culturally responsive counselling included within these provisions. Special attention was directed towards vulnerable groups, including LGBTQ+ individuals and first-generation students, to ensure fairness and inclusivity in the provision of support.

University-level interventions aimed at enhancing work–life balance and health behaviours among academic staff were recommended for systematic adoption. Policies such as flexible working hours, mental health leave, and workload management protocols were endorsed, as their effectiveness in reducing burnout and promoting well-being has been evidenced (Clark, 2000). Wellness initiatives for students were proposed to include access to nutritious food options, exercise facilities, and routine health screenings, measures shown to strengthen mental health outcomes and improve academic performance (Keyes, 2002). To enhance transferability, concrete practices such as faculty–student mentorship schemes, inclusive peer-support clubs, and digital well-being platforms were suggested as direct applications of the findings. Institutional decision-making was advised to be guided by robust data, with validated tools, including the model developed in this study, identified as essential for monitoring well-being trends and tailoring interventions to specific needs. Through the institutionalisation of such policies and the embedding of accountability at leadership levels, universities were positioned to transition from reactive mental health responses towards proactive, system-wide well-being strategies. These approaches were aligned with global objectives for sustainable and inclusive education (Waleewong & Yueayai, 2022).

4.4. Potential Limitations and Directions for Future Research

Several limitations were acknowledged in this study despite its theoretical and practical contributions. First, the cross-sectional design constrained the capacity to infer causal relationships among the constructs of social participation, health literacy, work–life balance, health behaviour, and mental health. Longitudinal research was therefore recommended to capture the dynamics of these variables over time, particularly in relation to institutional reforms or life-stage transitions. Second, although the sample size (n = 600) was sufficient for confirmatory factor analysis, recruitment from selected higher education institutions limited the generalisability of the findings. Cultural, institutional, and geographical variations were recognised as potential factors shaping the manifestation and interpretation of well-being constructs across diverse academic contexts. Third, the inclusion of both academic staff and students within a single model may have contributed to the emergence of dimensions, such as work–life balance, that are less directly applicable to the student population. This issue highlighted the need for population-specific refinements in future research. Finally, reliance on self-reported online questionnaires introduced potential sources of bias, particularly social desirability and recall bias, which may have reduced the accuracy of responses in sensitive domains such as mental health.

Future research was recommended to build on these findings by employing multi-method and longitudinal approaches, integrating qualitative interviews, behavioural tracking, or physiological measures to deepen understanding. The structural differences observed between staff and students should be further examined to explore role-specific pressures, coping mechanisms, and support systems influencing well-being in academic environments. Interventions based on the validated model, such as targeted programmes to enhance social participation or health literacy, should be implemented and their impact assessed over time. Expansion of the study into international or cross-cultural contexts was advised to provide insights into the universality or contextual variability of the proposed model. Finally, incorporation of digital health technologies and institutional policy variables into future frameworks was suggested to open new avenues for strengthening health and mental well-being in higher education.

5. Conclusions

The primary objective of the study was defined as the establishment of a psychometrically robust framework capable of capturing the multidimensional nature of well-being in higher education contexts. A quantitative research design was employed, with data collected through online and in-person questionnaires administered between September 2024 and February 2025. The final sample consisted of 600 participants, including undergraduate, graduate, and doctoral students, together with academic and support staff, recruited from a range of Thai universities to reflect institutional and regional diversity. Rigorous psychometric testing, incorporating both exploratory and confirmatory factor analysis, was undertaken and demonstrated strong reliability, validity, and structural coherence across student and staff populations. Confirmatory factor analysis produced satisfactory structural model fit indices (RMSEA = 0.047, CFI > 0.90), thereby confirming the integrity of the model across groups. Theoretical perspectives from public health, psychology, and education were integrated to underpin the framework, resulting in an empirically grounded tool suitable for future research, programme design, and policy development. The findings highlighted the importance of tailoring institutional support mechanisms to the distinct needs of different academic roles and provided actionable insights for advancing holistic well-being, equity, and resilience within the evolving landscape of higher education.