Synchronous and asynchronous online learning

Synchronous online learning refers to situations in which students and instructors meet in real time despite being geographically separated [9]. This mode of learning enables immediate interaction and feedback similar to face-to-face instruction [10], facilitated through tools such as video and audio conferencing, chat, web cameras, and other real-time communication technologies [11], and students can also participate in asynchronous engagement through tools such as discussion boards. Syn enables students who experience feelings of social disconnection in online environments to engage more actively in learning [9] and receive real-time feedback, which has been shown to positively influence learning satisfaction [8] and persistence [12]. However, other researchers [9] have noted that the instructional benefits commonly attributed to Syn depend substantially on the degree of interaction incorporated into the session, with non-interactive, lecture-centered formats demonstrating comparatively limited effects on learner engagement and affective outcomes. Although Syn had not been widely adopted as a formal component of conventional online learning, it proliferated rapidly during the pandemic under the label of Emergency Remote Teaching (ERT). Yet this rapid expansion was also accompanied by criticism, particularly regarding insufficient preparation and the absence of the structured instructional design that characterizes high-quality online learning [1–3].

In contrast, Asyn is defined by its reliance on asynchronous interaction, where communication and engagement occur at different times [9,13,14]. To support this mode of learning, instructors are encouraged to provide diverse online learning materials, such as pre-recorded videos, that learners can access anytime and anywhere [15,16]. A key feature of Asyn is delayed communication, which allows participants to interact and respond at various intervals [11]. This approach enables learners to complete tasks remotely with internet access within flexible timeframes, such as a weekly schedule. Zhang et al. [8] summarized the characteristics of Asyn, stating that it encompasses activities such as watching pre-recorded lessons, viewing slide presentations, or engaging in discussions at varied times.

Building on this distinction, prior research has examined various dimensions of online learning and implemented multiple interventions aimed at improving learning performance. In general, studies have reported that Syn offers advantages related to presence, interaction, and learner satisfaction [8,12], whereas Asyn has been associated with increased motivation in collaborative writing [17], opportunities for repeated engagement with lesson materials at one’s own pace, and flexible participation anytime and anywhere [15]. However, the research outcomes pertaining to the comparative effectiveness of Syn and Asyn reveal inherent complexity. For example, Nieuwoudt [4] highlighted the critical role of fostering meaningful interactions in online education, emphasizing the importance of providing abundant opportunities and options for interaction, as the pivotal contrast between Syn and Asyn predominantly lies in their approaches to interaction. However, Iglesias-Pradas et al. [18] indicated that the choice of Syn and Asyn and that of virtual communication tools did not significantly affect students’ academic performance.

Taken together, the relative strengths and limitations of Syn and Asyn appear to vary depending on factors such as the learning context, instructional conditions, and the ways in which instructors design and implement each modality. Consequently, neither mode can be unequivocally prioritized over the other. These findings underscore the ongoing need for approaches and research aimed at mitigating the weaknesses and amplifying the strengths of both modalities. In this regard, it remains worthwhile to further investigate the effectiveness of these two forms of online learning. Against this backdrop, examining COC which are purposefully designed through systematic instructional processes becomes critical for understanding how well-prepared online learning environments may address the limitations observed in both Syn and Asyn.

Conventional online courses (COC)

COC constitute a qualitatively different online learning modality, as they are deliberately designed through comprehensive instructional design processes that extend beyond the rapid or minimally structured implementations often associated with Syn and Asyn [1]. In practice, COC are implemented in at least two distinct forms: institutionally developed online courses that undergo systematic instructional design and development, and pre-developed OER-based courses that are incorporated into institutional curricula with varying degrees of pedagogical adaptation. Prior research has characterized institutionally developed COC as being intentionally designed through the integration of technological and pedagogical innovations to meet high-quality instructional standards [3]. This institutional form underscores the necessity for instructors to undergo a comparatively extensive and intricate instructional design process, ensuring the delivery of a quality learning experience that surpasses instructors’ self-created content. Hodges et al. [1] explained that a full-course development project can take months as it includes full-course design support, professional development opportunities, content development, LMS training and support, and multimedia creation in partnership with faculty experts.

On the other hand, due to the difficulty of producing a sufficient number of high-quality COC, higher education institutions have attempted to effectively conduct educational activities by utilizing pre-developed online courses or open educational resource (OER) repositories, such as OpenCourseWare (OCW) and Massive Open Online Courses (MOOCs). The OCW model has served as a catalyst for expanding access to high-quality university learning resources through open dissemination, thereby institutionalizing a culture of openly shared knowledge that aligns with the broader aims of OER to advance a more equitable and globally shared educational commons [19]. In fact, significant efforts and research have been dedicated to meaningfully integrating open educational resources such as OCW into educational settings. For example, studies have explored the importance of perceived administrative support in promoting the use of OCW [20], as well as how OCW can be introduced and effectively utilized within higher education institutions [19]. However, despite the widespread implementation of online learning in higher education settings, studies investigating the incorporation of pre-developed online courses such as OCW as learning materials remain scarce, emphasizing the need for relevant studies.

Taken together, although Syn, Asyn, and COC share core characteristics as online learning modalities, they differ substantially in their instructional design and implementation processes within higher education institutions. Accordingly, understanding how learners perceive the relative value of each modality is essential for deriving evidence-based implications for the effective design and sustainable implementation of online learning environments.

Extended information system success model (ISSM)

The effectiveness of a newly introduced information system in an organization must be validated. In an effort to demonstrate this, Delone and McLean [21] laid the groundwork for theoretical research by proposing a research model that comprehensively analysed prior studies pertaining to the success of information systems. They suggested that further development and validation are needed for their model [22]. Several researchers have since modified and tested the DeLone and McLean [21] model. For example, Seddon [23] and Rai et al. [24] asserted the necessity of incorporating “Perceived Usefulness (PU)” into the model as they considered “System Use” to be a behavior. However, DeLone and McLean’s [25] revised information system success model (ISSM), which selectively integrated diverse research findings and outcomes from prior studies on information system success, excluded PU.

Subsequently, researchers introduced additional variables to the Delone and McLean [25] model to investigate the influence of specific factors on user satisfaction and the intention to sustain usage. Wang [22] drew attention to the ongoing debate regarding the concepts of use and PU in ISSM, presenting an alternative model that once again incorporates the PU concept. Puspitarini and Ardhani [26] also contended that the updated DeLone and McLean ISSM cannot be entirely accounted for by the three technical aspects influencing it. This is because the “Information Technology Use” factor involves elements from a psychological decision perspective, thereby indicating the need to include psychological variables. According to Sabeh et al.’s [27] research on the inclusion of additional theoretical variables/key factors in IS success factor studies, there was a significant utilization of user-related factors (e.g., SE) and system quality-related factors (e.g., PU) within the original/updated Delone and McLean IS success model.

Efforts have been ongoing to elucidate the application of diverse technologies, grounded in ISSM, while simultaneously considering various exogenous variables. For example, studies examining university students’ continued intention to participate in online learning revealed that PU played a critical role in the success of information systems, such as satisfaction and CI [28]. Other studies (e.g., [29,30]) enhanced the ISSM model by introducing a SE construct as a precursor to user satisfaction and usage [31]. Kishore and McLean [32] argued that SE influenced PU in the context of technology adoption. They also explained that prior to technology adoption and use, people may use their SE as an anchor to form beliefs about the usefulness outcomes from technology adoption. Islam et al. [33] showed that PU mediated the relationships between SE, satisfaction, and intention to use technology when teachers adopted new technology.

One of the most critical issues for understanding successful online learning is teaching presence, which reflects the instructional, facilitative, and organizational roles of the instructor [13]. Teaching presence emerged in the Community of Inquiry framework [34], which was designed to support the design, development, and assessment of effective online learning [35]. The role of teaching presence has recently received significant attention, especially in the online learning environment. Teaching presence describes students’ perception of the accessibility, professionalism, and supportiveness of teachers [10]. Studies have reported that teaching presence significantly influenced the satisfaction [36] and persistence [37] of online learners, and it was the most crucial variable in determining course outcomes [38]. Furthermore, Zhang et al. [8] posited that an elevated level of teaching presence correlated with improved performance in Syn compared to Asyn, indicating that online learning formats are also subject to such influences. Therefore, to comprehensively evaluate learners’ perceptions across Syn, Asyn, and COC, it is necessary to integrate the technological perspective of ISSM with the pedagogical perspective of TP. Considering these discussions, the model in this study extends the original ISSM by incorporating key psychological factors (SE, TP) alongside the traditional system factors of SQ and IQ. This extended framework integrates technological and psychological dimensions, providing a comprehensive foundation for examining learners’ perceptions across different online learning modalities. Given the distinct instructional characteristics of the three online learning modalities, this enriched ISSM enables a systematic comparison of how modality-specific features influence PU, LS, and CI. Specifically, Syn emphasizes real-time interaction [11], rendering teaching presence particularly salient [10], whereas Asyn supports learner autonomy and are more closely associated with SE [8]. In contrast, COC, which are systematically designed with structured multimedia pathways, are more likely to enhance IQ and PU, thereby supporting instructional effectiveness [1,3] (Fig 1). Given that service quality refers to the quality of support services to system users [39], and that all participants in this study used the same LMS system within a single institutional context, service quality was not included as a separate construct in the present model.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Research model.

https://doi.org/10.1371/journal.pone.0345955.g001

Despite the substantial growth of online education and related research, studies specifically examining college-level students’ CI within the context of higher education remain relatively limited. In particular, empirical evidence regarding how the distinct characteristics of Syn, Asyn, and COC differentially influence learners’ intention to continue usage is scarce. Consequently, this study adopts the ISSM—widely recognized as a well-established framework for examining system adoption and continuance behavior—as the primary theoretical lens to compare these learning modalities. Clarifying these mechanisms through the extended ISSM is essential for understanding how learners respond to diverse online learning environments. Therefore, the present study aims to elucidate the structural relationships underlying these three online learning types by applying an extended ISSM, thereby enhancing our understanding of learners’ perceptions and CI across Syn, Asyn, and COC.

Participants

This study analyzed responses from 795 undergraduate students (male, 403, 50.7%; female, 392, 49.3%) enrolled at a university in South Korea. Over a two-week period in mid-December 2022, the survey was distributed via the Learning Management System (LMS) of specific courses that were pre-identified as either Syn, Asyn, or COC. To ensure the validity of the modality classification and minimize potential confusion arising from learners’ concurrent enrollment in multiple courses, participants were explicitly instructed to respond to the survey items based solely on their learning experience in the specific course where they accessed the survey link. Among the respondents, 313 students reported experience with Syn, 316 with Asyn, and 166 with COC.

This study involved a minimal-risk, non-interventional survey conducted as part of regular educational practices. In accordance with national legislation and guidelines in South Korea—including the Bioethics and Safety Act, which allows exemptions for anonymous, practice-based educational research that does not involve identifiable or sensitive personal information—the study was determined to be exempt from formal Institutional Review Board (IRB) review. Therefore, IRB approval was not required. All procedures adhered to ethical standards for research involving human participants, and no personally identifiable or sensitive data were collected. All participants were informed about the purpose of the study, and participation was voluntary and based on informed consent.

Research instruments and measures

To examine the association and the difference between the types of online learning, a survey questionnaire consisting of seven study variables and demographic questions was used. The survey questions were adapted from previous studies (Table 1). The seven validated measures used in the research model were as follows: system quality (SQ), information quality (IQ), academic self-efficacy (SE), teaching presence (TP), perceived usefulness (PU), learning satisfaction (LS), and continuance intention to use (CI).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Measures.

https://doi.org/10.1371/journal.pone.0345955.t001

SQ refers to the information system’s usability, accessibility, utility, complexity, and response time [25]. In this study, accessibility, stability, response time, and ease of use are considered to be the important aspects of a good system. IQ was described by Delone and McLean as the level of output produced by an information system and is a crucial aspect of determining the efficacy of information [47]. Questionnaire items on IQ included those regarding system usefulness, understandability, interestingness, reliability, completeness, timeliness. SE refers to the subjective belief in one’s ability to successfully learn or perform given learning tasks [48], which encompasses “self-efficacy for learning” and “self-efficacy for performance.” TP was assessed using 15 items, encompassing systematic lesson execution by teachers, facilitation and assessment of learning, as well as the concept of instructor presence. PU was evaluated using items that inquired about the improvement of learners’ learning outcomes, and effectiveness and efficiency regarding online learning. LS was assessed using questions on learning quantity and value, and whether respondents encouraged others to take the same online course. CI included questionnaire items on preference, continuous intention, and expanding and recommending online learning.

The 48 items in the questionnaire were measured on a 5-point Likert scale. As all respondents were Korean, the original survey items written in English were first translated into Korean by researchers, then reviewed by three bilingual PhDs in educational technology to ensure content validity. The Cronbach’s alpha of the scale was 0.97 in the present study.

Data analysis

To comprehensively examine the differences between Syn, Asyn, and COC, this study adopted a sequential analytic framework spanning measurement invariance, latent mean analysis (LMA), and multi-group structural equation modelling (SEM). This approach was chosen to ensure a rigorous comparison: first, measurement invariance tests confirmed that constructs were conceptually equivalent across groups; next, LMA identified differences in the magnitude of learners’ perceptions; and finally, multi-group SEM revealed variations in the underlying structural mechanisms. This integrated strategy allows for simultaneously examining both the quantitative levels and qualitative pathways distinguishing the three types. Prior to these main analyses, normality tests were conducted to ensure that the data met the necessary assumptions. To make cross-group comparisons, the latent mean analysis approach was used for latent variables that could not be directly measured. This approach has been recommended for cross-group comparisons while the variables are conceptualized as latent constructs that could not be directly measured [49].

Measurement invariance is a prerequisite for making cross-group comparisons using latent mean analysis [49]. Before testing the group differences in the means of latent variables, the configural, metric, and scalar invariance must be examined to evaluate the equivalence of the groups [49]. Finally, measurement invariance tests were conducted to determine latent mean differences across the three groups [50–52].

Next, a structural model invariance test was employed across the groups to assess the significance of differences in each path. This involved comparing a series of models with different constrained paths to determine if the responses across the groups were equivalent. Testing the structural model invariance across the groups involved two main procedures: metric invariance equality constraints [53], which examined the differences in paths between the groups. As a precondition to compare the structural path coefficient differences, metric invariance equality tests were conducted in subsequent steps using the same baseline research model. This process involved conducting a test of the unconstrained model followed by the equal factor loading model and the equal path model.

SEM was used to examine the hypothetical model and was employed using maximum likelihood estimation to test the measurement model’s fitness. Analyses were performed using AMOS version 21. χ2 test was employed to evaluate the model fit. To evaluate model fit, we used the Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), and Root Mean Square Error of Approximation (RMSEA). Following established guidelines, CFI and TLI values greater than 0.90 were considered indicative of a good fit [54]. For RMSEA, values less than 0.05 indicate a close fit, while values between 0.05 and 0.08 are considered reasonable errors of approximation [55].

Correlation, means and standard deviations

Table 2 presents the correlations and descriptive statistics of the seven measured variables, which were significantly positive correlated (r = .343 to.822, p < .001). Both convergent validity and discriminant validity were satisfied [56]. Univariate normality was confirmed based on skewness values ranging from −0.617 to −0.108 and kurtosis values ranging from −0.571 to 0.447, all falling within the recommended thresholds of |3| and |10|, respectively [56].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Correlation, means and standard deviations (N = 795).

https://doi.org/10.1371/journal.pone.0345955.t002

Latent mean analysis

Several invariance tests were performed before conducting the latent mean analysis. The baseline configural, metric, and scalar invariance models all fit well to the data (Table 3). First, the configural invariance was assessed without constraining equality across the groups (M1). The goodness-of-fit results (χ2 = 1107.49, CFI = 0.959, TLI = 0.948, RMSEA = 0.039) indicated similar structural patterns across the groups. This implies that the configural model can serve as a baseline for comparison with other restricted models in the invariance tests. Subsequently, metric invariance was assessed by constraining the factor loadings to be equal across the groups (M2). The results demonstrated a good model fit (χ2 = 1153.276, CFI = 0.957, TLI = 0.949, RMSEA = 0.038). Finally, a scalar invariance test was performed by constraining the intercepts across the groups to be invariant (M3). The model fit indices showed that χ2 = 1553.906, CFI = 0.938, TLI = 0.926, and RMSEA = 0.046. Given that the assumptions of configural, metric, and scalar invariance were satisfied; ΔCFI<0.02 [50], ΔTLI<0.05 [51], ΔRMSEA< 0.015 [52], a latent mean analysis could be conducted.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Fit indices for invariance verification.

https://doi.org/10.1371/journal.pone.0345955.t003

Latent mean analysis was employed to estimate the mean differences across the three groups. COC were selected as the primary reference group (fixed to zero [49]), as COC represents the standard of online instruction designed with systematic instructional principles, serving as a baseline for comparing the distinct characteristics of Syn and Asyn. Furthermore, to comprehensively examine the differences between all three modality pairs (including Syn vs. Asyn), we systematically rotated the reference group. This procedure allowed us to obtain precise estimates and statistical significance for the mean differences among all three groups, as presented in Table 4. Overall, COC had higher mean values for all factors than Syn and higher mean values for six variables than Asyn except for system quality. Asyn had higher mean values for SQ, SE, and CI than Syn, whereas Syn had higher mean values for TP than Asyn.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Results of latent means analyses.

https://doi.org/10.1371/journal.pone.0345955.t004

Structural model invariance test across the groups

The structural model invariance test across the groups assesses whether the proposed theoretical model is parameter invariant across groups. The SEM of the hypothesized model on the whole sample was performed before conducting the multi-group analysis. The results of the SEM showed that the proposed model for the whole model yielded a good fit, χ2 = 776.792, CFI = .961, TLI = .953, RMSEA = .066 (Table 5). Subsequently, the multi-group analysis structural equation model analysis was conducted.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Summary statistics for tested models in multi-group analyses.

https://doi.org/10.1371/journal.pone.0345955.t005

The structural model invariance across the groups was analyzed by testing the metric invariance equality constraints, which examined whether the responses across the groups were identical in the research model, and the cross-group equality constraints, which examined whether there were differences in the paths across the groups. This enabled us to determine whether the samples from the three groups yielded consistent results within the same research model and to analyze any differences exhibited by the research model across the groups.

First, all parameters in the model were released among the groups, and M1 was obtained (χ2 = 1188.074, CFI = 0.954, TLI = 0.945, RMSEA = 0.04). Second, the factor loads were equalized among the groups and M2 was obtained (χ2 = 1236.435, CFI = 0.953, TLI = 0.946, RMSEA = 0.04). Lastly, the paths between the latent variables were equalized among the groups, and M3 was obtained (χ2 = 1276.593, CFI = 0.951, TLI = 0.946, RMSEA = 0.04). Subsequently, M2 and M1, M3 and M2 were compared (Table 5). The fit statistics for all three models were acceptable, as they consistently satisfied the recommended benchmark criteria (CFI, TLI > 0.90; RMSEA < 0.08). When the fit statistics of M2 and M1 were compared, it was found that they did not change significantly (ΔCFI = 0.001, ΔTLI = −0.001, ΔRMSEA = 0). Finally, analysis of the change in the fit statistics of M3 and M2 revealed that the change of the fit values were not significant (ΔCFI = .002, ΔTLI = 0, ΔRMSEA = 0). Therefore, the paths in the model were valid for the three types of online learning.

Second, cross-group equality constraints were tested to examine the significant differences in the relationships among the variables of the research model across the groups. Nine models, each imposing equality constraints on the path coefficients within the model, were compared with the baseline model that did not impose equality constraints on any path coefficients (Table 6). The results revealed four significant path differences in the model between the groups. They also indicated that the paths from IQ to PU (P2; Δχ2 = 9.334, p = 0.009), IQ to LS (P6; Δχ2 = 7.236, p = 0.027), PU to LS (P7; Δχ2 = 14.47, p = 0.001), and PU to CI (P9; Δχ2 = 7.03, p = 0.03) were significantly different between the three groups.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Fit indices for different models with constraints on each path.

https://doi.org/10.1371/journal.pone.0345955.t006

Table 7 shows the standardized regression weights of the paths for each type of online learning. Consistently, the paths from IQ to LS (P6), PU to LS (P7), LS to CI (P8), and PU to CI (P9) exhibited significant influence across all three groups. However, SQ did not have a significant influence on PU (P1) and LS (P5). In contrast, the paths from IQ (P2), SE (P3), and TP (P4) to PU exhibited different degrees of influence across the three groups. After testing cross-group equality constraints, the results showed that the effects of the paths in the model differed across the three groups (Fig 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. Path coefficients of the multi-group analyses.

https://doi.org/10.1371/journal.pone.0345955.t007

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. The standardized coefficients for path with the significant differences between groups.

Note. * p < .05; ** p < .01; *** p < .001.

https://doi.org/10.1371/journal.pone.0345955.g002

Latent mean differences across modalities

Structural path invariance (Multi-group analysis) across modalities

To move beyond mean-level differences and examine whether the underlying ISSM mechanisms operated similarly across modalities, a multi-group structural equation model was conducted. A few statistically significant differences were observed between the types of online learning according to the model tested with multi-group analysis. The path coefficients between IQ and LS, PU and CI, PU and LS, and IQ and PU showed significant differences. The relationship between IQ and LS, PU and CI, and PU and LS were thus significant in all three models, whereas that between IQ and PU was not significant in the Syn model.

Firstly, the coefficients between IQ and LS showed that COC was the highest, followed by Asyn and Syn (COC = 0.493, Asyn = 0.309, Syn = 0.157). This implies that the level or presentation of information provided in Syn may not have adequately met learners’ demands. Previous research has emphasized the importance of IQ—such as providing students with sufficient, concise, and clear information, as well as updated and engaging learning content—in enhancing overall LS and CI in online learning environments [58]. Additionally, the delivery method or structure of information in Syn may not have sufficiently contributed to learners’ overall satisfaction [59]. From an ISSM perspective, this finding suggests that in learning environments where instructional explanations are primarily delivered through pre-structured content—as is typical in COC—IQ plays a central role in shaping learner satisfaction. In contrast, in Syn, limitations in IQ may be less salient due to the presence of immediate instructor interaction and real-time clarification.

Second, the relationship between PU and CI also showed that COC coefficients were significantly higher than Syn and Asyn. This suggests that PU plays a more influential role in shaping CI within the COC. Consequently, to support online learning continuity, instructional design approaches characteristic of COC may enhance learners’ perceptions of online learning usefulness [1,3]. This underscores the importance of clear and well-structured instructional information in strengthening learners’ PU and supporting CI in COC environments.

However, the relationship between PU and LS exhibited coefficients opposite to those observed in the PU to CI relationship. Specifically, Syn ranked the highest, followed by Asyn and COC. These contrasting patterns imply that the function of PU differs by modality. In COC, the structured and well-designed nature of the course environment strengthens the ISSM mechanism in which PU directly promotes CI. In contrast, the Syn setting, characterized by real-time interaction and instructional immediacy, reinforces the experiential component of learning, thereby amplifying the influence of usefulness on LS rather than on CI. Such findings highlight the necessity for ongoing research in online education on how to enhance meaningful interaction and ensure learning satisfaction for students.

In the Syn model, the relationship between IQ and PU was not significant, whereas the paths in the models of Asyn and COC were. This implies the need to enhance IQ in future Syn learning environments. Higher IQ is crucial for learners to perceive the usefulness of online learning, thereby increasing LS and their CI [47]. Moreover, in terms of IQ, the quality of information provided in well-designed COC is likely to be higher than that in Asyn, where instructional design processes may be less systematically applied.

According to the ISSM framework, learners’ CI is shaped by a sequential mechanism in which quality perceptions (e.g., IQ, SQ) and psychological variables (e.g., TP as an extended factor) influence PU, which in turn affects LS and ultimately CI. The patterns observed in this study should therefore be interpreted not merely as descriptive differences but as modality-dependent variations in the operation of the underlying ISSM mechanism across online learning formats.

In addition to the significant structural paths, several non-significant or relatively weaker relationships merit brief consideration. These patterns may reflect modality-specific instructional characteristics rather than theoretical inadequacy. For example, in Syn environments, real-time interaction and instructor immediacy may reduce learners’ reliance on structured informational cues, thereby attenuating the influence of IQ on PU. Conversely, TP significantly influenced PU only in the Syn model, highlighting the unique value of live instructor facilitation. In addition, the non-significant influence of SQ across all groups, potentially coupled with ceiling effects in highly structured COC, may have constrained variability in certain system-related perceptions. Such findings suggest that the operation of ISSM mechanisms is contingent on instructional modality and contextual design features, underscoring the importance of modality-sensitive interpretation rather than uniform expectations across online learning formats.