2.1 Performance expectancy, effort expectancy, social influence, and facilitating conditions

The UTAUT model, proposed by Venkatesh et al. (2003), has been widely used in various fields, such as business, education, and communication. It includes four core variables: PE, EE, SI, and FC, with four moderators: age, gender, experience, and voluntariness. Several studies have confirmed the positive effects of the four core variables on BI in the context of e-learning (Wang et al., 2009). Among the constructs explaining MOOC usage intention, PE and EE are the most commonly studied in the literature (Pozon-Lopez et al., 2020). PE reflects a learner's belief in the benefits of using MOOCs for learning or job performance, while EE refers to the perception of ease in participating in MOOCs. MOOC adoption and usage are more likely among learners who have a high perception of PE and find the system easy to use (Alasmari & Zhang, 2019; Ameen et al., 2019).

SI refers to a learner's perception of how much their significant others (e.g., classmates, peers, friends, or teachers) believe they should use MOOCs, and these opinions can influence their intention to engage in an e-learning system (Bere, 2014). FC relate to an individual's perception of the technology, technical infrastructure, tools, system, and support available when using MOOCs. Venkatesh et al. (2003) found that there was a positive association between FC and UB (Venkatesh et al., 2012). Moreover, Zhou et al. (2019) demonstrated that SI and FC were significant predictors of BI in the context of hospital electronic information management system use, and FC also has a significant effect on intention and UB. Therefore, we propose the following hypotheses:

2.2 Perceived cost, perceived pleasure, and self-regulated learning management

While the UTAUT model provides a strong foundation for understanding technology acceptance, we propose expanding it to better capture the unique aspects of MOOC adoption and usage. Specifically, we incorporate three new constructs, namely PC, PP, and SRL management (SM). PC refers to a learner's PC of using MOOCs, such as infrastructure costs, time, and effort to access content, and of changing a learning tradition or habit. We include PC to account for the potential barriers to MOOC adoption. While many MOOCs are free or low-cost, learners still incur nonmonetary costs such as time and effort (Flake et al., 2015; Liu et al., 2024). Understanding how learners perceive these costs is crucial for predicting MOOC adoption, especially given the high dropout rates often observed in MOOCs. By including PC, we can gain insights into how learners weigh the benefits against the costs when deciding to use MOOCs (Douglas et al., 2020; Yang et al., 2019). Flake et al. (2015) proposed that increased costs lead to negative appraisals of effort. Therefore, we hypothesize the following:

PP is an important factor to consider in MOOC adoption as the voluntary nature of most MOOCs means that learners' intrinsic motivation plays a significant role. A pleasurable experience has been shown in previous studies to be a predictor of satisfaction with MOOCs (Pozon-Lopez et al., 2020). Users are more likely to be intrinsically motivated to adopt new technology if they find it enjoyable, fun, and pleasurable (Lee, 2010; Wang et al., 2009), and if it provides them with a sense of achievement or enhances their knowledge and skills (Douglas et al., 2020). Including PP allows us to examine how the hedonic and motivational aspects of MOOCs impact adoption and usage. Therefore, the following hypothesis is proposed:

Finally, SRL is crucial in the MOOC context due to the learner-centered and self-paced nature of these courses (Onah et al., 2024; Zimmerman, 2002). Unlike traditional classroom environments, MOOCs require learners to independently manage their learning process, set goals, and monitor their progress (Liu et al., 2024; Pozon-Lopez et al., 2020). We incorporate the construct of SM to better understand how learners' ability to self-regulate impacts their intention to use and the actual usage of MOOCs. We hypothesize that learners with strong self-regulation skills are more likely to feel confident in their ability to successfully engage with MOOCs, potentially increasing their intention to use these platforms (7a.). Furthermore, effective self-regulation can help learners overcome challenges and persist in their MOOC studies, directly influencing their UB (7b.). Therefore, the following hypotheses are proposed in this study:

2.3 Behavioral intention and its mediating role

The relationship between BI and UB has been well-established in technology acceptance research. Numerous studies have consistently demonstrated that BI is a strong predictor of actual UB across various technological contexts (Bere, 2014; Lee, 2006). This relationship has been further corroborated in the specific domain of e-learning systems, where intention has been shown to positively influence the actual system use (Ameen et al., 2019; Mohammadi, 2015; Venkatesh et al., 2003).

Beyond its direct influence, BI also serves a crucial mediating role in the adoption and use of technology. Research has highlighted its significance as an intermediary factor that bridges the gap between various antecedents and actual UB (Lee, 2006; Zhou et al., 2010). A notable example of this mediating effect can be found in the study by Zhou et al. (2019), which revealed that BI significantly mediated the influence of various factors on nurses' adoption and use of Health Information Exchange Management Systems.

In the context of MOOCs, this study proposes that BI plays a dual role—both as an antecedent to UB and as a dependent variable influenced by FC and SM. FC, which encompass factors such as resource availability, technical support, and infrastructure can significantly shape a learner's intention to use MOOCs. Similarly, an individual's SM skills, including goal setting, strategic planning, and time management, are likely to influence their intention to engage with MOOCs. These factors, mediated through BI, are expected to ultimately impact learners' actual UB toward MOOCs.

Based on this theoretical foundation and previous empirical evidence, we propose the following hypotheses:

2.4 Moderating roles of gender, grade, major, and experience

Previous research has shown that factors such as age, gender, major, and experience can moderate the relationship between BI and various factors (Taylor & Todd, 1995; Venkatesh & Bala, 2008). In this study, to better align with the characteristics of MOOC learners, the age variable was adjusted to grade level. Considering that the individual UB may be related to their majors, we took major as an important moderator. In fact, experience is commonly used as a moderator in technology acceptance research (Venkatesh & Bala, 2008), as knowledge gained from previous behaviors can influence users' intentions. In sum, the moderating variables of gender, grade, major, and learners' experience in MOOCs (LEX) were incorporated into the research framework to increase the predictive validity of the model.

Based on the above discussions and assumptions, we established a model to further understand the drivers of BI and UB toward MOOC learning. The model considered a range of factors that were deemed relevant to the topic (Figure 1).

3.1 Instrumentation

The questionnaire utilized in this study was adapted from previous research to suit the context of MOOCs. It consisted of 38 items written in Chinese, measured on a 5-point Likert-type scale ranging from “1 = strongly disagree” to “5 = strongly agree.” The questionnaire comprised nine constructs, including PE, EE, SI, FC, PC, PP, SM, BI, and UB. Four items were used for PE, EE, FC and SI each (Venkatesh et al., 2003), three items for PC (Jin-Hong et al., 2015), four items for PP (Lee, 2010; Wang et al., 2009), five items for SM (Wang et al., 2009), and four items for both BI and UB (Lee, 2006; Venkatesh et al., 2003). Prior to the formal survey, a pilot survey was conducted, and reliability and factor analyses were performed on the pilot data. As a result, two items were dropped from the questionnaire due to validity and reliability concerns, leaving a total of 36 measurement items for the final survey. The survey also collected sociodemographic information such as grade, gender, major, and MOOCs experience of the learners. Detailed information and reference sources for the scales and items used in the questionnaire can be found in Appendix A.

3.2 Data collection and participants

The study focused on Chinese college students who had prior learning experience in MOOCs and utilized a convenience sampling method to collect data. Data collection was conducted using a blend of offline and online methods. Offline data were collected from participants across 10 universities in southeast China. Online data collection was facilitated through a professional online survey platform, which ensured the anonymity of respondents and made the process of participation more convenient. Before administering the survey, the purpose of the study was explained to the participants and their voluntary consent was obtained. Confidentiality and protection of personal information were assured. A total of 602 questionnaires were collected, out of which 548 were considered valid for data analysis. Responses from 54 individuals were omitted due to incompleteness and lack of experience in MOOCs. The characteristics of the respondents are summarized in Table 1.

4.1 Measurement model

In this study, the internal reliability of the scale was assessed using Cronbach's alpha coefficient, which resulted in a high reliability score of 0.963. The reliability of each construct was evaluated by examining the factor load, with Cronbach's alpha values ranging from 0.764 to 0.911 for all constructs, surpassing the recommended threshold of 0.7. These results indicated that the scale demonstrated good structural and referent validity.

The content validity of the scale was ensured as most of the measurement items were derived from established scales in the existing literature. Exploratory factor analysis results indicated an acceptable structure, Kaiser–Meyer–Olkin = 0.965, chi-square = 12,953.83, df = 630, and p = 0.000. These findings suggested that the scale possessed adequate construct validity and was suitable for factor analysis.

Convergent validity of the scale was evaluated using factor load, comprehensive reliability (CR), and average variance extracted (AVE). The factor load of each item ranged from 0.656 to 0.861, surpassing the recommended threshold of 0.6. The CR values for each factor ranged from 0.764 to 0.911, exceeding the acceptable level of 0.7, indicating consistent explanation of each construct by its items. The AVE values for all constructs ranged from 0.506 to 0.718, higher than the lower bound of 0.50, with CR values always greater than AVE values, indicating favorable conditions for convergent validity. All constructs in the model exceeded the threshold recommended by Hair et al. (2010), indicating good convergent validity. Detailed information on the reliability and validity of the measurement model can be found in Table 2.

Table 3 presents the results of the correlation analysis, indicating a statistically significant relationship (p < 0.01) between each construct. This finding provides further evidence that the items are designed to measure the same construct. Additionally, the diagonal elements in the correlation matrix were replaced by the square root of the AVE value for each construct. The AVE values, which were >0.5 for each construct, were significantly higher than the correlations with other constructs. These results demonstrate that the data met the conditions for discriminant validity and indicate good discriminant validity.

4.2 Structural model for hypotheses testing

The results of the confirmatory factor analysis, as presented in Table 4, revealed that the correlation coefficients between the measured items and their corresponding constructs were statistically significant (p < 0.05). However, in the initial model, the negative influence of PC on BI was found to be insignificant (ß = −0.017, p = 0.609). To ensure a more parsimonious model with a good fit to the data, PC was removed from the model. This resulted in refined model-fitting indices that indicated a good degree of fit, as all the values exceeded the recommended thresholds (Hair et al., 2010). The analysis of variance of the regression model also showed that the regression equation was significant.

The results of hypothesis testing for the standardized path coefficients and path significance are presented in Table 5. With the exception of 5., most of the paths showed significant results in the expected direction. Specifically, there were significant positive relationships between BI and PE (ß = 0.228, p < 0.01), EE (ß = 0.165, p < 0.05), SI (ß = 0.157, p < 0.05), PP (ß = 0.138, p < 0.05), FC (ß = 0.337, p < 0.01), and SM (ß = 0.188, p < 0.01). This supports hypotheses 1., 2., 3., 4a., 6., and 7a.. Additionally, there were significant positive relationships between UB and BI (ß = 0.252, p < 0.01), FC (ß = 0.337, p < 0.01), and SM (ß = 0.146, p < 0.01), supporting hypotheses 4b., 7b., and 8..

4.3 Mediation model

The mediation analysis was conducted using the causal steps method (Baron & Kenny, 1986) and the regression coefficients were tested using structural equation modeling. The results revealed several significant findings: (1) FC (ß = 0.188, p = 0.003) and SM (ß = 0.146, p = 0.029) had significant direct effects on BI; (2) BI had a significant direct effect on UB (ß = 0.252, p = 0.000); and (3) the effects of FC (ß = 0.337, p = 0.000) and SM (ß = 0.330, p = 0.000) on UB were also significant. As a result, hypotheses 9a. and 9b. are supported, indicating that BI acts as a partial mediator. The total effect of FC and SM on UB was 0.337 and 0.330, respectively, with the mediation effect accounting for 12.3% and 10.0% of the total effect, respectively.

4.4 Moderation model

To examine the moderating roles of gender, grade, major, and LEX differences on college students' online learning, we conducted multi-group analyses. In both analyses, each pathway coefficient was constrained to be equal across the two gender, grade, major, or LEX groups. The subsamples were grouped and compared using AMOS24.0, according to the moderators. The results of the fitting indexes were changed, as presented in Table 6.

The results of the group comparison revealed that there were no significant differences in gender, grade, and major between the constrained and unconstrained models. The regression coefficient difference was also insignificant (p > 0.05), and the critical ratio of chi-square to the degree of freedom (p > 0.05) did not show any significant changes.

Furthermore, the LEX was considered as a moderator, and the ratio of minimum sample difference to the degree of freedom was 2.77 (CMIN = 24.900, df = 9, p = 0.003). In the next stage, a multi-linear regression analysis was conducted to test the moderating role of LEX. All data were centralized, and the moderation analysis was performed. The model and regression coefficients, calculated using SPSS 24.0, are presented in Table 7.

Moreover, the findings revealed that the LEX moderated the positive effects of EE on BI (F = 5.173, p = 0.023), as well as the effects of SM on UB (F = 8.057, p = 0.005). However, LEX had no significant effect on the relationship between PE, SI, PC, FC, PP, SM, and BI, as well as the relationship between FC and BI and UB.

The moderating effects of LEX on the relationship between EE and BI were further examined by plotting the simple regression slopes at low and high levels of LEX (mean ± 1 standard deviation). The results revealed a significant and positive correlation between EE and BI at both low and high levels of LEX. However, the slope of the relationship was steeper for learners with low LEX compared to those with high LEX, as depicted in Figure 2. Specifically, when LEX was at a low (high) level, an increase in the probability of BI from −0.842 to 0.85 (−0.538 to 0.502) was observed. This means that the impact of EE on BI is more pronounced for learners with lower levels of experience. With an increase in LEX, learners gradually adapt to the use of the system, and the external cognitive load of using the system decreases correspondingly; thus, the influence of EE on BI decreases accordingly.

In Figure 3, the slope of a low LEX is much steeper than that of a high LEX. As SM decreases from one standard deviation above the mean to one standard deviation below, the probability of UB increases significantly faster when learner's MOOCs experience (LEX) is low (from −1.061 to 0.953) than when it is high (from −0.569 to 0.637). A possible explanation is that less experienced learners begin to form SM ability in an MOOC environment, resulting in SM having a more significant impact on UB.

4.5 Results of the model

The standardized path coefficients and variances explained by the model are presented in Figure 4. The empirical results provide support for all hypotheses, except for 5.. Specifically, the results indicate that BI is significantly influenced by six exogenous variables, with an R2 of 0.802. FC and SM have both direct and indirect effects on UB through BI, resulting in an R2 value of 0.729.

5.1 Performance expectancy, effort expectancy, social influence, and facilitating conditions

Our findings on the significant positive influences of PE, EE, and FC on BI align with previous research on technology acceptance in online learning contexts. For instance, Ameen et al. (2019) and Wang et al. (2009) found similar relationships in their studies of e-learning systems. Moreover, Alasmari and Zhang (2019) conducted a mixed-methods study with 1203 Saudi Arabian college students, extending the UTAUT model to examine mobile learning acceptance. While their focus was on mobile learning rather than on MOOCs specifically, they similarly found that factors such as learning expectancy and EE significantly predicted usage intentions. This consistency across studies reinforces the robustness of these factors in predicting MOOC adoption (Yu et al., 2024). The significant role of SI in our model echoes findings by Mohammadi (2015) and Fianu et al. (2020), who highlighted the importance of peer and instructor support in online learning environments. Our results suggest that the social aspects of learning remain crucial even in the seemingly individualized MOOC format. This underscores the need for MOOC designers to incorporate social learning features and for institutions to create supportive cultures around MOOC usage.

Additionally, we observed a significant positive correlation between FC and UB in our study. The extent to which learners utilize MOOCs is influenced by the FC that surround their usage. When resources are easily accessible and conditions are conducive to the learning process, learners are more likely to use MOOCs frequently. Therefore, it is crucial to design MOOCs with a learner-friendly interface and functionality, especially catering to less experienced learners, so that technical issues do not hinder their learning experience (Lee, 2006). MOOC providers should prioritize usability features by providing learner support, intuitive interfaces, and efficient content delivery. Timely support and frequently asked questions (FAQs) can aid learners in searching for course materials and engaging in learning activities.

5.2 Perceived pleasure and perceived cost

Our study's findings align with and expand upon previous research regarding the role of PP and PC in MOOC adoption. For instance, Wu and Chen (2017) used structural equation modeling with a sample of 252 Chinese MOOC users to integrate the TAM and TTF model. Their results showed that perceived usefulness and attitude were critical to continuance intention, which aligns with our findings on the importance of PP. It is worth noting that our results indicate that PP outperforms traditional UTAUT constructs in predicting BI toward MOOCs. This underscores the importance of creating an enjoyable and engaging learning experience to attract and retain MOOC users. MOOC designers and developers should prioritize creating media-rich, interactive environments that stimulate learner interest and satisfaction. Incorporating gamification techniques and other interest-generating elements could enhance the intrinsic rewards of MOOC participation, potentially improving user engagement and retention (Reparaz et al., 2020).

Regarding PC, our study contributes to a relatively understudied area in MOOC research. While cost analysis has not been extensively examined in the context of MOOCs (Flake et al., 2015), existing literature presents mixed findings on its impact on BI (Douglas et al., 2020). Our results indicate a negative influence of PC on BI, though this relationship was not statistically significant. This unexpected outcome may be attributed to the current MOOC landscape, characterized by abundant free or low-cost options and minimal switching costs for learners. Specifically, the proliferation of MOOC projects by universities and governments worldwide has led to widespread availability of diverse course resources. Furthermore, the ubiquity of mobile devices has eliminated the need for learners to invest in additional hardware for MOOC participation. These factors collectively reduce both financial and nonfinancial costs (such as time and effort) associated with MOOC adoption, potentially explaining why PC did not emerge as a significant predictor in our model.

5.3 Self-regulated learning management

Results revealed that self-regulation capacity is crucial for predicting users' BI and UB toward MOOCs. Recent research by Onah et al. (2024) highlights that MOOC learners often lack specific SRL skills, particularly in help-seeking and task strategies. Even learners with strong formal education backgrounds may struggle to apply effective SRL strategies in MOOCs. To support diverse learner needs, MOOCs should provide tools for goal setting, strategic planning, and self-monitoring (Li & Lajoie, 2022; Zhu et al., 2020). Examples include embedding learning support tools, displaying progress bars, and setting clear tasks with deadlines (Onah et al., 2024). These features enable learners to monitor their activities, utilize social tools for self-regulation, and adjust their strategies based on timely feedback. Practically, MOOC platforms can better support learners in developing and applying necessary self-regulation skills by incorporating SRL-enhancing features. These SRL-enhancing features could potentially improve engagement, persistence, and learning outcomes in MOOCs (de Barba et al., 2020; Zheng et al., 2023; Zimmerman, 2002).

5.4 Mediating effects of behavioral intention

The intention-behavior gap suggests that not all BI translates into actual behavior. However, the results of our study support the positive impact of BI on UB, which is consistent with the findings of Lee (2006) regarding the use of e-learning systems. This study further confirms the importance of SM, FC, and BI on UB, consistent with previous MOOC studies (Fianu et al., 2020; Gregori et al., 2018; Wu & Chen, 2017). Moreover, our findings indicate that PE, EE, SI, and PP—all of which have significant effects—positively impact learners' UB, indirectly through BI. This aligns with the results of previous studies that identify BI as a significant predictor and mediator of UB in e-learning systems (Ameen et al., 2019; Khan et al., 2018; Mohammadi, 2015). In particular, FC and SM significantly impact learners' UB, and their influence is mediated by BI. This implies that manipulating factors such as FC and SM can potentially increase learners' BI and consequently promote their MOOC UBs. Additionally, improving learners' PP of using MOOCs can also positively influence their UBs. Therefore, course designers, instructors, and teachers should design MOOC environments that enable learners to perceive the usefulness, ease of use, and pleasure of MOOC learning.

5.5 Moderating effects of experience

Previous research has shown that prior experience with online learning plays a significant role in predicting the adoption and persistence of MOOCs (Semenova, 2016). The TAM has also highlighted the moderating role of experience in shaping BI and UB toward information systems (Zheng & Li, 2020). In this study, we conducted multi-group analyses and found that the learner's experience with MOOCs moderated the relationship between EE and BI, as well as between SM and UB. This finding is consistent with previous research by Taylor and Todd (1995), which showed that the positive effect of perceived ease of use is higher for less experienced learners. Furthermore, our findings align with the research by Venkatesh and Bala (2008), who found that the impact of perceived ease of use on BI decreases over time as users gain more experience with the system.

Learners' experience mediates the relationship between SM and UB. It is highly possible that learners with more experience may have developed better SRL strategies and habits, leading to more effective management of their learning activities, and ultimately influencing their UB. On the other hand, less experienced learners may require more guidance and support in developing SRL skills, which can affect their management of learning tasks and subsequently influence their UB. Therefore, it is important to provide instructions to help novice learners develop the necessary SRL skills for learning in MOOCs (Yang et al., 2019). Cultivating their SRL skills and habits, along with providing guidance from facilitators, can help overcome initial obstacles (Yang et al., 2019; Zhu et al., 2020). Course providers and instructors should also tailor their instruction to the learner's experience level, providing more support for novice learners to facilitate their SRL processes.

5.6 Theoretical and practical implications

The theoretical implications of this study are multifaceted. First, by integrating PC, PP, and SRL into the UTAUT model, we provide a more comprehensive framework for understanding MOOC adoption that accounts for both external factors and internal motivations. This study addresses calls in the literature for more holistic models of technology acceptance in education (Khan et al., 2018; Yu et al., 2024). Second, our findings on the moderating role of MOOC experience contribute to the ongoing discussion about how user characteristics influence technology adoption processes (Liu et al., 2024; Venkatesh & Bala, 2008). This highlights the need for future research to consider user experience levels when studying MOOC usage patterns.

From a practical standpoint, our results offer several implications for MOOC providers, instructors, and educational institutions. The strong influence of PP suggests that MOOC designers should focus on creating engaging, enjoyable learning experiences. This could involve incorporating gamification elements, interactive content, and opportunities for creative expression. Additionally, the significance of SM points to the need for scaffolding and support tools to help learners develop these crucial skills (Onah et al., 2024). MOOC platforms could integrate features like customizable study planners, progress trackers, and self-reflection prompts to facilitate better self-regulation. Furthermore, our findings on the moderating effects of MOOC experience suggest that providers should tailor their interfaces and support systems to users' familiarity levels. For instance, novice users might benefit from more extensive onboarding processes and simplified interfaces, while experienced users could be offered advanced features and more complex learning pathways. By addressing these implications, MOOC providers can create more effective, engaging learning environments that cater to a diverse user base and potentially improve completion rates and learning outcomes.