Social cognitive theory (SCT) was first put forward by Bandura (1986), which explains that individuals’ behavior is not only affected by environmental factors, but also by personal factors, presenting a triadic reciprocal determinism. The theory emphasizes that there is an interactive and interdependent relationship among person, environment and behavior(Hong et al., 2015; Uçar and Sungur, 2017), which is widely used to identify the influencing factors of individual behavior in various specified fields (Chiu and Tsai, 2014; Schunk and DiBenedetto, 2020; Zhou et al., 2020), such as information science (Middleton et al., 2019), career development (Gregory and Penela, 2023), and education (Hust et al., 2017).

The current study included the three factors of SCT model that have been found to be significantly vigorous for predicting learning outcomes, specifically personal factors, environmental factors and behavioral factors. In the learning investigation context, behavioral factors include not only learning styles such as networked learning and digital learning, but also learning outcomes, as they involve the actual activities and actions undertaken by individuals (Lee, 2020). The emergence and development of behavior change are affected by personal factors, for instance, digital literacy and self-efficacy for digital technology, which are activated when individuals possess digital literacy and feel capable of achieving desired behavior (Lee and Tseng, 2024). Moreover, environmental factors play an indispensable and important role in the learning investigation context, which include not only the physical environment, but also the social and learning environment that individuals are in every day. Digital atmosphere, emerging with the digital age, forms a vital part of this learning environment. Thus, we incorporated the combination of personal, environmental and behavioral factors in our study for the purpose of providing a comprehensive framework for understanding the relationship between digital literacy and learning outcomes among college students.

Digital literacy was first mentioned by Gilster (1997), which was used as “the ability to both understand and use digitized information in multiple formats from a wide range of sources when it is presented via computers.” The definition of digital literacy has not yet reached a consensus in academia (Wei, 2024), and there are several perspectives on the definition of digital literacy in the literature. Buckingham (2015) stated that digital literacy refers to various functional issues related to computers and information technology, not just of a functional issue of computers. Polizzi (2020) defined the digital literacy as the ability to access, analyze, evaluate and produce messages in a variety of forms. In a more global context, digital literacy is defined by UNESCO (2018) as the ability to access, manage, understand, integrate, communicate, evaluate and create information safely and appropriately through digital technologies for employment, decent jobs and entrepreneurship, which includes competences that are variously referred to as computer literacy, ICT literacy, information literacy and media literacy. Researchers have established key literacies of digital literacy and the specific literacies identified varied according to the nature of the studies conducted (Falloon, 2020). For instance, Eshet (2004) developed a conceptual framework for digital literacy, which consisted of five literacies, including (1) photo-visual literacy, (2) reproduction literacy, (3) information literacy, (4) branching literacy, and (5) socio-emotional literacy. Based on the work of Eshet (2004), Ng (2012) offered a conceptual framework viewing digital literacy as emerging at the convergence of students’ technical, cognitive, and socio-emotional competencies, which focuses on an individual’s objective ability to solve practical problems in digital scenarios.

Atmosphere no longer denotes only the envelope of gases around a planet, but in contemporary usage has become a metaphor for the emotional field generated when people engage with particular settings or events (Paschou and Papaioannou, 2023). For instance, digital atmosphere of museums was characterized by the emotions elicited by the integration of digital applications, as well as elements such as content, design, color choices, sound and audio, and the scents and aromas within the physical space (Paschou and Papaioannou, 2023). Visiers-Jiménez et al. (2021) propose that the pedagogical atmosphere on the ward has been identified as part of what constitutes a high-quality clinical learning environment. Virtanen et al. (2022) note that the PE students felt that the learning atmosphere of their final year study module was warm and conversational according to the research results. Tucker and Goodings (2017) report that atmosphere is commonly understood as the feeling or sense of a particular space, and introduce digital atmosphere as a lens to foreground the affective currents that shape how people seek and offer care within social media spaces. In this study, digital atmosphere, born out of the digital age, refers to the perceived feeling or sense that students experience within digital learning spaces, drawing on the concept of learning atmosphere (Virtanen et al., 2022) and educational climate (Krupat et al., 2017) and extending it to digital contexts. Digital atmosphere includes dimensions such as perceived support, interactivity, and emotional tone within digital learning environments. According to the social cognitive theory (SCT), an individual’s level of digital literacy may affect their use and perception of digital technology, which in turn influences the digital atmosphere they are in (Steelman et al., 2016). The first hypothesis of the study is as follows:

With the widespread adoption of digital technology and artificial intelligence in the education landscapes, digital learning has come into being but does not alter the fundamental nature of education (Ismail and Nur, 2024). In the digital age, learners need to not only acquire traditional skills, but also develop a comprehensive ability to comprehend, assess, synthesize, utilize, and generate knowledge by leveraging digital technologies. This is crucial for obtaining professional knowledge, boosting professional competencies, and ensuring the conditions and support necessary for continuous learning, lifelong learning, and innovation. According to the social cognitive theory (SCT), digital atmosphere, as an environmental factor, may impact digital learning behavior of college students. The second hypothesis of the study is formulated as follows:

Self-efficacy was defined by Bandura (1986) as the belief in one’s ability to exert control in a given atmosphere and to produce a desired outcome. This belief is also understood as the confidence in applying one’s skills to achieve a specific result (Yang and Du, 2024). Individuals with high perceived self-efficacy are more likely to be motivated to take action and have a greater willingness to tackle challenging tasks (Gerdts-Andresen et al., 2023). Self-efficacy for digital technology refers to one’s perception of their ability to perform technological tasks or use digital technology to handle tasks (Yuen-Han Mo et al., 2025), which is a type of “ability belief” that emphasizes an individual’s subjective perception of their own abilities, rather than the abilities themselves. In the context of social cognitive theory, the higher an individual’s level of digital literacy, the more likely they are to have greater self-efficacy in using digital technology. What’s more, perceived self-efficacy for digital technology may influence college students’ digital learning. Getenet et al. (2024) found that positive student attitudes and digital literacy significantly contributed to self-efficacy, which, in turn, positively affected the engagement dimensions. Yang and Du (2024) pointed that students with higher perceived self-efficacy for digital technology are more likely to actively use digital tools for learning. Thus, the third and fourth hypotheses of the study are as follows:

The learning and teaching process carried out in educational institutions is reflected in the achievement of optimal learning outcomes. The way students achieve learning outcomes is demonstrated through the lively discussion and interaction conducted by both teachers and students. The learning outcome has become an integral and inseparable aspect for assessing the quality of education (Puniatmaja et al., 2024). Learning outcomes are typically understood as the results that emerge from the interactions that take place during the learning process. Lin et al. (2017) found that digital learning shows better positive effects on learning outcome than traditional teaching does. Hsiao et al. (2019) confirmed that students’ online learning behaviors had a significant effect on short-term learning outcomes, and indicated that students with a higher frequency of online learning behaviors obtained a higher GPA grade level. The fifth hypothesis of the study is deduced:

An individual’s level of digital literacy may influence the digital atmosphere they are in Steelman et al. (2016). Additionally, digital atmosphere, as an environmental factor, may impact digital learning behavior of college students. Through digital learning, college students may achieve better learning outcomes by using digital technologies and tools. Therefore, we argue that digital atmosphere, self-efficacy for digital technology and digital learning play mediating roles in the relationship between digital literacy and learning outcomes. The sixth and seventh hypotheses of the study are as follows:

The relationship description model is designed to clarify the nature of the relationships between two or more variables, as well as the degree of their correlation, while also identifying and quantifying covariance (Karasar, 2012). In this study, we adopt this model and utilize structural equation modeling (SEM) techniques to conduct an in - depth analysis. Specifically, we explore the interplay and correlations among digital literacy (DL), digital atmosphere (DA), self - efficacy for digital technology (SEDT), digital learning (DL’), and learning outcomes (LO) among college students, as depicted in Figure 1.

As illustrated in Figure 1, path H1 represents the direct effect of digital literacy on digital atmosphere; path H2 represents the direct effect of digital atmosphere on digital learning; path H3 represents the direct effect of digital literacy on self - efficacy for digital technology; path H4 represents the direct effect of self - efficacy for digital technology on digital learning; and path H5 represents the direct effect of digital learning on learning outcomes. All analyses were conducted using SPSS 21.0, the PROCESS macro in SPSS, and AMOS 21.0, with statistical significance set at α = 0.05.

The present study utilized a questionnaire survey methodology. The survey was administered via the online platform from October to November 2024, targeting college students in universities located in the western region of China. A total of 589 individuals participated in the survey, representing a diverse range of academic disciplines. After excluding invalid questionnaires, a sample size of 538 was retained, comprising 47.96% female participants (n = 258) and 52.04% male participants (n = 280). In terms of sample size, Maxwell et al. (2008) highlighted the significance of sample size planning in influencing statistical power and the precision of parameter estimates. Lakens and Ravenzwaaij (2022) underscored the necessity of justifying sample size adequacy to ensure the robustness of research findings. Consistent with the recommendations of Schumacker and Lomax (2010), numerous studies have employed sample sizes ranging from 250 to 500 participants, which is deemed sufficient for conducting structural equation modeling (SEM). Given these considerations, the sample size in the current study is deemed adequate.

To ensure data quality and identify inattentive respondents, we incorporated two validity screening measures in the questionnaire design. We included two instructional manipulation checks (IMCs):(1)A simple arithmetic verification item, “What is the answer to 3 + 5?”; (2) A semantic recognition task that asks participants to identify the non-fruit item from a list containing apple, banana, stone, and orange. Additionally, we implemented an automated response time tracking system to record completion durations. Responses that failed either attention check (incorrect answers to verification items) or had implausibly short response times (below 15 s, as determined by pilot testing percentiles) were excluded from analysis. Through this rigorous validation protocol, 51 invalid responses (9.5% of total submissions) were eliminated, resulting in 538 validated questionnaires retained for formal analysis. This validation rate of 90.5% exceeds recommended thresholds for online survey research, ensuring sufficient statistical power for subsequent analyses. Table 1 presents the sample data statistics. 52.04% of the participants were male, and 86.25% of the participants were majoring in liberal arts, science, and engineering.

This study utilizes a 5 - point Likert scale, with options spanning from “strongly agree” to “strongly disagree.” It is divided into five dimensions, namely digital literacy, digital atmosphere, self-efficacy for digital technology, digital learning, and learning outcomes. In the process of adapting and designing the scales, we meticulously translated each item into Chinese and subsequently back-translated them into English. This procedure was instrumental in verifying the precision and consistency of the items across both linguistic and cultural spectrums. Beyond this, we were particularly attentive to ensuring the contextual relevance of the items to the Chinese educational landscape. To achieve this, we carefully revised the wording to align with local educational nuances and consulted with Chinese - system - savvy educators. Their insights ensured the adapted scales were linguistically accurate, contextually appropriate, and culturally resonant, thus enhancing their validity and applicability in China.

After the collection of the scale data, we carried out data analysis with the aid of tools of SPSS 21.0 and AMOS 21.0.

The data analysis procedure comprised seven sequential steps to ensure methodological rigor. First, we conducted a common method bias assessment using Harman’s single-factor test to examine potential homogeneity bias in self-reported measures. Second, reliability analysis was performed through Cronbach’s alpha coefficient calculation for all multi-item scales to evaluate internal consistency. Third, competing measurement models were compared using confirmatory factor analysis (CFA) with maximum likelihood estimation to identify the optimal factorial structure. Subsequently, we evaluated the measurement model fit through multiple indices (χ²/df, CFI, TLI, and RMSEA) and assessed construct validity by examining both convergent validity. Fifth, descriptive statistics (means and standard deviations) were computed for all study variables, followed by bivariate correlation analysis using Pearson’s coefficients to establish preliminary relationships among constructs.

Finally, mediation analyses were conducted through bias-corrected bootstrapping procedures (5,000 resamples) to examine indirect effects between key dimensions, controlling for relevant covariates. All analyses were performed using SPSS 21.0 and AMOS 21.0 with statistical significance set at α = 0.05.

In the assessment of the measurement model, the fit between observed and latent variables and sample data was analyzed. The measurement model was tested using confirmatory factor analysis (CFA) to check the validity and reliability of the items explaining the endogenous variables. The goodness of fit of the model was evaluated against several criteria, including X²/df, CFI, TLI, RMSEA, and SRMR. A measurement model is considered to have an acceptable fit if X²/df < 5, CFI > 0.90, TLI > 0.90, RMSEA < 0.08, and SRMR < 0.08. Based on these criteria, the measurement model of the present study received an acceptable fit (X²/df = 3.177, CFI = 0.929, TLI = 0.920, RMSEA = 0.064, SRMR = 0.037). These results indicate that the measurement model adequately represents the data, with the X²/df ratio well below 5, CFI and TLI values exceeding 0.90, and RMSEA and SRMR values below the recommended thresholds, suggesting a reasonable approximation of the data by the model.

Moreover, the study also calculated average variance extracted (AVE) and composite reliability (CR) for convergent validity and reliability of all constructs. The recommended limits for AVE and CR are 0.5 (Fornell and Larcker, 1981) and 0.7 (Nunnally and Bernstein, 1994) respectively. The AVE values for all constructs exceeded 0.5, and CR values surpassed 0.7, as shown in Table 2. These results confirm that the majority of variance in the items can be attributed to the underlying constructs, and that the measures are reliable. For instance, the AVE for digital literacy was 0.588, indicating that 58.8% of the variance in the items is explained by the digital literacy construct.

Reliability analysis was performed using Cronbach Alpha and composite reliability (CR) coefficients. Cronbach Alpha and composite reliability coefficients indicate the reliability of the measures, since both should be over 0.70 (Hair et al., 2022). The scales measuring the three variables are found out to be valid and reliable.

Furthermore, the items with weak factor loadings were deleted to improve the AVE and overall fit of the model. A discriminant validity analysis was also conducted, and all values suggested acceptable discriminant validity. The results of the discriminant validity analysis, presented in Table 3, indicate that each construct is empirically distinct from the other constructs in the model. For example, the square root of the AVE for digital literacy (0.767) is greater than its correlations with all other constructs, confirming that digital literacy is distinct from digital atmosphere, self-efficacy for digital technology, digital learning, and learning outcomes.

The AVE and CR values are provided in Table 2, and values obtained as a result of discriminant validity analysis are provided in Table 3. These results collectively demonstrate that the measurement model exhibits sufficient validity and reliability, providing a robust foundation for the subsequent structural equation modeling analysis.

In this study, Pearson correlation analysis was conducted using SPSS 21.0 and AMOS 21.0 to examine the relationships among multiple variables: digital literacy (DL), digital atmosphere (DA), self-efficacy for digital technology (SEDT), digital learning (DL’), and learning outcomes (LO). All scales used were 5-point Likert-type scales. As shown in Table 4, the mean values of these five variables were all above the scale midpoint.

The Pearson correlation analysis revealed significant positive correlations between digital literacy, self-efficacy for digital technology, digital atmosphere, digital learning, and learning outcomes, as indicated in Table 4.

All five hypothesized direct relationships in the structural model received empirical support. Specifically, digital literacy exhibited a significant and positive influence on the development of a digital atmosphere (β = 0.589, p < 0.001) and students’ self-efficacy for digital technology (β = 0.826, p < 0.001). Additionally, a well-developed digital atmosphere positively impacted digital learning (β = 0.830, p < 0.001), and higher self-efficacy for digital technology significantly enhanced digital learning (β = 0.317, p < 0.001). Digital learning was strongly associated with positive learning outcomes (β = 0.833, p < 0.001). The β coefficients and corresponding bootstrap confidence intervals are presented in Table 5. The structural model with hypothesis testing results is shown in Figure 2.

Moderation analysis was performed using the Bootstrap method at 95% confidence level. 10,000 different indirect effects are calculated by selecting 10,000 different samples from the sample in the Bootstrap method. These effects are used to determine the lower and upper limits of the 95% confidence interval (CI). As long as there is no zero (0) value between the lower (LLCI) and upper (ULCI) values, the relationships in the model are significant (Hayes, 2017).

The analysis of indirect pathways confirmed some of the proposed hypotheses (Table 6). Digital literacy exhibited a positive impact on learning outcomes through digital atmosphere and digital learning (β = 0.407), lending support to Hypothesis 6. Additionally, digital literacy positively influenced learning outcomes via the self-efficacy in digital technology and its subsequent effect on digital learning (β = 0.218), thereby supporting Hypothesis 7.

Although this study provides valuable insights for college students and universities in the digital age, it also has several limitations that should be addressed in future research. One limitation is the potential for social desirability bias inherent in survey-based research. Respondents may be influenced by social expectations and choose answers that align with social norms or mainstream opinions rather than reflecting their true thoughts (Rosenthal, 2002). To enhance the representativeness of the sample, future studies could employ a variety of data collection methods to triangulate the results.

Additionally, this study focuses on the impact of digital atmosphere and self-efficacy for digital technology on digital learning and learning outcomes, based on social cognitive theory. However, in the context of digital education, there are likely other influencing factors at play. Future research should explore these additional factors to gain a more comprehensive understanding of the landscape. Conceptually, despite maintained statistical boundaries, the proximity between digital literacy and self-efficacy for digital technology remains a focus for refinement in future iterations. The concept of “digital atmosphere” is relatively new, and there is no consensus on its definition in the academic community. There is potential for further research on this topic in the future.

The generalizability of the results may be limited to the digital infrastructure and educational ecosystem within Chinese higher education. To extend the applicability of these findings, further cross - cultural validation is necessary.

Lastly, the primary subjects of this study are college students, who are primarily in the role of “learners” in the educational process. However, education is a two-way street, and teachers and universities also play crucial roles. To mitigate the limitations of self-report bias and sample homogeneity, future investigations may utilize longitudinal or experimental designs to substantiate causal relationships. Integrating teachers’ insights would enhance student-centered approaches and offer a more comprehensive perspective on the digital learning environment. Moreover, examining domain-specific impacts (e.g., STEM versus humanities) or cultural factors influencing digital self-efficacy could significantly enrich the depth of our comprehension in this field.