A survey was done to examine the behavioral intentions to eLearning adoption among faculty members departments of nursing institutions in Kuwait. The period for data collection was between from September to November 2024.

Conveniently, 65 nursing faculty members in nursing institutions in Kuwait. Participants had varying qualifications and utilized eLearning that encompassed the diverse views of faculty members engaged in nursing education were included in the study. Completion of the survey tool implied consent to voluntarily participate in this study.

A survey was done to examine the behavioral intentions to eLearning adoption among faculty members departments of nursing institutions in Kuwait. Specifically, the questionnaire employed established Likert-type scales, ranging from 1 (Strongly Disagree) to 5 (Strongly Agree), which are commonly used in Technology Acceptance Model (TAM)-based research. The items were adapted and modified from validated instruments in prior studies (e.g., Davis, 1989; Venkatesh and Davis, 1996; Teo and Noyes, 2011; Al-alak and Alnawas, 2011), ensuring theoretical alignment and content validity for constructs such as Perceived Usefulness (PU), Performance Expectancy (PE), Behavioral Intention (BI), Digital Literacy (DL), and Information Literacy (IL).

Additionally, a pilot test was conducted with 12 nursing faculty members to evaluate the clarity and relevance of the instrument in the Kuwaiti academic context. Feedback from the pilot led to minor wording adjustments for clarity.

In the main study, reliability was assessed using Cronbach’s alpha and Composite Reliability (CR), with all constructs exceeding the recommended threshold of 0.70. Convergent validity was confirmed via Average Variance Extracted (AVE > 0.50), and discriminant validity was verified using both the Fornell-Larcker criterion and HTMT ratio, as detailed in Tables 1–3. These evaluations align with widely accepted standards in SEM analysis (Hair et al., 2017; Fornell and Larcker, 1981).

The statistical analysis was conducted using Structural Equation Modeling (SEM) with Partial Least Squares (PLS), an approach that is well-suited for exploratory studies involving complex models and smaller sample sizes. SEM allows for the simultaneous analysis of multiple dependent relationships among latent and observed variables. Data analysis was performed using Smart-PLS software, which supports robust PLS-SEM estimation. The outer measurement model was evaluated for reliability and validity, utilizing Cronbach’s alpha, Composite Reliability (CR), Average Variance Extracted (AVE), and discriminant validity measures, including the Fornell-Larcker criterion and Heterotrait-Monotrait (HTMT) ratio. Additionally, the Variance Inflation Factor (VIF) was calculated to check for multicollinearity among predictor variables. The reliability of constructs was confirmed through composite reliability and Cronbach’s alpha, with values exceeding the threshold of 0.70, while convergent validity was established with AVE values above 0.50. These values support the robustness of the measurement model, ensuring reliable and valid interpretation of the structural model’s relationships.

Path coefficients were assessed to examine direct and mediating effects among constructs. Bootstrapping with 5,000 resamples was employed to determine the significance of the paths, producing t-statistics and p-values for each relationship.

Ethical approval was obtained from the Research Ethic Committee of the Public Authority for Applied Education and Training, College of Nursing, State of Kuwait, (Ethical Approval No. CoN: 10/23). Written informed consent was obtained from each of the faculty members in nursing institutions in Kuwait prior to the study.

Figure 1 presents a simplified conceptual framework of the modified Technology Acceptance Model (TAM) used in this study. The model illustrates the relationships among perceived ease of use (PEOU), perceived usefulness (PU), attitude toward use, and behavioral intention, incorporating the added constructs according to the modifications. The relationships hypothesized within this framework were tested using Structural Equation Modeling (SEM) using Smart-PLS. SEM is a comprehensive statistical technique that enables the analysis of complex relationships among observed and latent variables, allowing for the testing of both direct and indirect effects (Bauldry, 2020; Shah and Goldstein, 2006).

The outer measurement model is used to assess the reliability, internal consistency, and validity of both observed (questionnaire-based) and latent variables (Ho, 2013). The results of measurement model analysis are shown in Table 1. To evaluate consistency, both Cronbach’s alpha and Composite Reliability (CR) are employed. Unlike Cronbach’s alpha, which assumes equal weighting of all items, Composite Reliability takes standardized loadings into account, providing a more precise measure of internal consistency (Fornell and Larcker, 1981). Discriminant validity is assessed through cross-loadings and the Fornell-Larcker criterion, helping ensure that constructs are distinct (Ali et al., 2018; Hair et al., 2017).

The items display mean values ranging from 3.5 to 4.2, suggesting a generally positive response across constructs. Moderate standard deviations (0.7–1.2) indicate some variability in participant responses, reflecting a favorable overall disposition toward the constructs under study and consistent support across items.

Outer loadings for all items surpass the recommended threshold of 0.70, demonstrating the reliability of these items as indicators of their respective constructs (Hair et al., 2017). High loadings are observed in items such as BI4 (0.92), PE2 (0.90), and PE3 (0.91), underscoring their strong representation of underlying constructs. Although a few items, such as IL4 (0.70), have slightly lower loadings, they remain within acceptable limits for exploratory analysis (Hair et al., 2010; Malhotra et al., 2006).

Multicollinearity was evaluated using Variance Inflation Factor (VIF) values, which range from 1.57 to 3.32, remaining well within the acceptable < 5 (Hair et al., 2010). This suggests that each item offers unique information for its construct, reinforcing the structural integrity of the model.

The reliability analysis, shown in Table 1, is based on CR and Cronbach’s alpha. Both measures exceed the accepted threshold of 0.7, indicating satisfactory reliability across constructs. For convergent validity, item loadings are expected to > 0.70, while Average Variance Extracted (AVE) values should be > 0.50 (Hair et al., 2010). As illustrated in Table 1, all factor loadings and AVE values meet these criteria, confirming the model’s convergent validity and further supporting the constructs for structural analysis (Table 1).

Discriminant validity was assessed to indicate that the construct is distinct from other constructs and accurately captures its intended variable (Sarstedt et al., 2014). To establish discriminant validity, a construct should correlate more strongly with its own indicators than with those of other constructs (Hair et al., 2017). In this study, discriminant validity was assessed using two approaches: The Fornell-Larcker criterion and the HTMT ratio.

The Fornell-Larcker criterion was used to assess the discriminant validity of the constructs in the measurement model (Table 2). This method compares the square root of the Average Variance Extracted (AVE) for each construct with the correlations between that construct and all other constructs. For discriminant validity to be confirmed, the square root of AVE (displayed on the diagonal) should be higher than any of the inter-construct correlations in the same row and column (Fornell and Larcker, 1981).

In this analysis, each construct’s square root of AVE exceeds its correlations with other constructs. These results confirm that each construct shares more variance with its items than with other constructs, satisfying the Fornell-Larcker criterion and indicating adequate discriminant validity in the model (Table 2).

The Heterotrait-Monotrait (HTMT) ratio of correlations was employed to assess discriminant validity among the constructs (Table 3). The HTMT approach, as proposed by Henseler et al. (2015), provides a stringent criterion for determining whether constructs in a model are sufficiently distinct. A widely accepted threshold is an HTMT value < 0.85. In this study, all construct pairs exhibit HTMT values below the conservative threshold of 0.85, indicating satisfactory discriminant validity (Table 3).

Structural model assessment is a crucial step in Structural Equation Modeling (SEM) that evaluates the relationships between latent variables in the hypothesized model. It involves examining the path coefficients for strength and significance. The results of the direct path analysis are shown in Table 5.

The path analysis reveals a range of significant relationships between constructs. Digital Literacy (DL) demonstrates a positive impact on Performance Expectancy (PE) (β = 0.36, t = 2.48, p = 0.013*), indicating that greater digital literacy corresponds with higher expectations for improved performance. The relationship between Digital Literacy and Feasibility of Adopting eLearning (F) is also positive, with a path coefficient of 0.21 (t = 1.93, p = 0.054), though this falls just above the threshold for statistical significance, suggesting a more limited association.

Feasibility of Adopting eLearning (F) significantly influences Behavioral Intention (BI), with a path coefficient of 0.34 (t = 2.06, p = 0.040*). This result implies that perceptions of eLearning feasibility contribute positively to individuals’ intentions to adopt eLearning technologies. Information Literacy (IL), however, does not show significant effects on either Feasibility of Adopting eLearning (β = 0.16, t = 1.37, p = 0.172) or Performance Expectancy (β = 0.18, t = 1.06, p = 0.288), indicating that information literacy may have a limited direct role in influencing these outcomes.

Perceived Usefulness (PU) exhibits a strong positive relationship with both Feasibility of Adopting eLearning (β = 0.26, t = 2.16, p = 0.031*) and Performance Expectancy (β = 0.38, t = 3.20, p = 0.001***). These significant results suggest that perceived usefulness is an influential predictor of both the feasibility of eLearning adoption and the expected performance outcomes.

Additionally, Performance Expectancy (PE) positively impacts Behavioral Intention (BI), as evidenced by a path coefficient of 0.36 (t = 2.89, p = 0.004**), which indicates that higher expectations of performance improvements enhance intentions to engage with eLearning. Performance Expectancy also significantly predicts Feasibility of Adopting eLearning (β = 0.32, t = 2.48, p = 0.013*), underscoring the importance of anticipated performance benefits in shaping perceptions of eLearning feasibility (Table 5).

The analysis of indirect effects reveals several significant pathways (Table 6). The pathway from Perceived Usefulness (PU) to Feasibility of Adopting eLearning (F), mediated through Performance Expectancy (PE), is significant with a path coefficient (β) of 0.12 (t = 1.99, p = 0.046*). This indicates that perceived usefulness indirectly enhances the feasibility of adopting eLearning by increasing performance expectancy. Similarly, Perceived Usefulness also shows a significant indirect effect on Behavioral Intention (BI) through Performance Expectancy, with a path coefficient of 0.14 (t = 1.97, p = 0.049*), suggesting that perceived usefulness positively influences behavioral intention via enhanced expectations of performance.

In contrast, several pathways did not achieve statistical significance. For instance, Information Literacy (IL) does not significantly affect Behavioral Intention (BI) through Feasibility of Adopting eLearning (F) (β = 0.05, t = 1.13, p = 0.260) or through Performance Expectancy (PE) and Feasibility (β = 0.02, t = 0.96, p = 0.34). The pathway from Digital Literacy (DL) to Behavioral Intention (BI) through Performance Expectancy (PE) and Feasibility (F) also fails to reach significance (β = 0.04, t = 1.16, p = 0.247). Additionally, Digital Literacy (DL) shows an indirect effect on Behavioral Intention (BI) through Performance Expectancy (PE) (β = 0.13, t = 1.93, p = 0.054) and Performance Expectancy to Behavioral Intention through Feasibility (β = 0.10, t = 1.65, p = 0.099), both approaching significance but not meeting the conventional threshold (p < 0.05) (Table 6).

The assessment of the measurement model demonstrates robust reliability and validity, underscoring the structural soundness of the constructs examined. Cronbach’s alpha and composite reliability exceed the acceptable thresholds, aligning with prior findings that emphasize the importance of internal consistency for effective e-learning model evaluations (Khan et al., 2023). High outer loadings further indicate that the constructs used in the study, such as perceived usefulness and behavioral intention, are well-represented, a key criterion for ensuring meaningful insights into technology adoption behaviors (Garrido-Gutiérrez et al., 2023).

However, the variability in certain item loadings, such as IL4, which marginally meets the threshold, suggests that specific aspects of the constructs may require refinement to better capture user perceptions. This aligns with Toprak and Aslan’s (Toprak and Aslan, 2020) assertion that construct development must account for contextual and cultural nuances, especially in cross-disciplinary studies. The use of the Fornell-Larcker criterion and the HTMT ratio to confirm discriminant validity demonstrates rigorous adherence to methodological standards, ensuring that the constructs measured are distinct and meaningful (Al-Maroof et al., 2021).

The assessment of the structural model reveals several significant relationships among the constructs, offering valuable insights into the factors influencing eLearning adoption.

The structural model assessment shows significant relationships among the constructs analyzed, providing valuable insights into the dynamics of eLearning adoption.

Perceived usefulness emerges as a critical predictor, significantly influencing both feasibility (β = 0.26, p = 0.031*) and performance expectancy (β = 0.38, p = 0.001*). This aligns with existing literature emphasizing perceived usefulness as a primary driver in technology acceptance models.

Digital literacy also shows a significant positive relationship with performance expectancy (β = 0.36, p = 0.013*), highlighting the important role of digital competencies in shaping teachers’ expectations of e-learning’s benefits. Although digital literacy shows only a marginal relationship with feasibility (β = 0.21, p = 0.054) suggests that digital literacy alone may not fully address the perceived barriers to e-learning, necessitating additional support mechanisms, such as user-friendly interfaces and targeted training programs. This finding resonates with research by Park and Shea (2020), which identified digital literacy as a pivotal factor in eLearning adoption.

In contrast, information literacy does not exhibit significant direct effects on feasibility or performance expectancy, diverging from expectations. This may imply that information literacy influences e-learning adoption indirectly or plays a less immediate role compared to digital literacy. As suggested by (Prior and Wilson, 2016), the impact of information literacy might be mediated through other constructs such as organization support or resource availability.

The structural model evaluation highlights several significant relationships, shedding light on the complex interplay of variables influencing e-learning adoption. For instance, the positive impact of digital literacy on performance expectancy aligns with the findings of Wijaya et al. (2022), who emphasize the foundational role of digital competence in shaping user expectations of technology. Similarly, the strong influence of perceived usefulness on both performance expectancy and behavioral intention underscores its pivotal role as a determinant of technology adoption, a pattern well-documented in studies using the UTAUT framework (Garrido-Gutiérrez et al., 2023; Khan et al., 2023).

Interestingly, the non-significant effects of information literacy on key outcomes, such as feasibility and performance expectancy, suggest that other factors may mediate these relationships. Al-Shamali et al. (2022) propose that organisational culture and resource availability often outweigh individual competencies in determining technology adoption, a perspective that could contextualize these findings. The significant indirect effects of perceived usefulness on behavioral intention, mediated through performance expectancy, also align with the broader literature, indicating that the perceived value of e-learning systems enhances user engagement (Al-Maroof et al., 2021).

The main limitation relates to the sample size of our study and therefore any generalization of the current study must be done with care. Secondly, conditions in which eLearning systems are employed may differ from country to country, and thus the styles and model in the study can be used in other studies, but the results obtained may be different from our study. Thirdly, not all the factors determining the relinquishment of eLearning systems were explored in the study, we recommend that nursing institutions develop a policy framework and a readiness checklist to facilitate the effective adoption of eLearning systems. Such a framework should include assessments of digital and information literacy levels among staff, infrastructure capacity, and organisational support mechanisms to enhance acceptance and utilization of eLearning tools. Additionally, future study will therefore examine the moderating impact of technological structure, cost of internet pack, and leadership commitment on the espousal of eLearning systems. Moreover, the study will also try to explore the agreement effect of perceived utility and perceived ease of use in the model estimated in this paper.