DC is a multi-faceted concept ( Sánchez-Caballé et al., 2020) that evolved from diverse backgrounds ( Gallardo-Echenique et al., 2015; Lucas, 2019). The UK higher education context proposed Digital Capabilities Framework having six elements ( Biggins et al., 2017) that can be used to enhance students’ ability to steer self-learning for continuous development. Likewise, the European Commission developed the DC framework (DigComp2.1) to respond to the ever-increasing need to operate effectively in a knowledge-intensive society ( Sillat, Tammets, & Laanpere, 2021). With five dimensions and 21 elementary competencies, this framework was first published for European citizens in 2013 and renewed in 2017. This framework highlights the significance of digital creation, innovation, communication, collaboration, engagement, and digital identity ( Lucas, 2019; Sillat et al., 2021). Later it was adopted within the education sector to create a standard for evaluating the DC of educators and students ( Lucas, 2019). Experts predict that acceleration in edutech growth will sustain, and DC training in higher education ( MHRD, 2020) will profoundly shift the focus towards using digital technologies to enhance students’ learning experiences and facilitate the development of their DC.

Regrettably, in a traditional learning environment, similar instruction styles are followed regardless of the individual learning abilities of students. The digital resources are designed at baseline, ignoring individual learners’ present DC levels ( Martzoukou et al., 2020). As students belong to different demographics, the requirement of levels of support for DC may vary ( Martzoukou et al., 2020). The diversity in socio-demographic characteristics may widen the digital divide ( Moore et al., 2018). Hence, it cannot be presumed that all students arrive at university with the same levels of DC. Some studies suggest that students develop DC spontaneously in digital learning environments through active engagement and self-motivation ( Heidari et al., 2021; Lucas, 2019; McGuinness & Fulton, 2019). At the same time, few others emphasize the close linkage between well-founded pedagogy, didactics, and DC ( Sung et al., 2016; Tamim et al., 2011). In the digital learning environment, it is argued that meaningful learning occurs when students are active, constructive, intentional, authentic, and cooperative ( Howland et al., 2012). The above standpoints deliberated by researchers with diverse backgrounds invite the inquiry of learning processes from students’ perspectives ( Blau et al., 2020).

Theoretically PL consists of cognitive, emotional, and social aspects that deal with understanding new insights, feelings and experiences during learning and inter-personal interactions through the learning sessions ( Blau et al., 2020; Rockinson-Szapkiw et al., 2016). It primarily relates to two predominant aspects of learning: knowledge acquisition and knowledge transfer ( Barbera et al., 2013) which are projected to be essential to acquire DCs. However, the prediction of DC having a significant relationship with PL has largely remained unexplored. There is no evidence thus far investigating this relationship in the extant literature related to online education. Hence, we propose the following research hypothesis:

H1: There is a significant positive relationship between students’ Digital competence and perceived learning in an online learning environment.

The researchers in the field of digital literacy and competence feel that mere usage of digital tools will not automatically make students digitally competent ( González & Martín, 2017; Sánchez-Caballé et al., 2020). There is a gap between formal (e.g. educational software, technology theory) and informal (e.g. multimedia tools) digital skills and abilities of university students ( Flores & Roig, 2016; Parvathamma & Pattar, 2013; Prieto et al., 2020; Purushothaman, 2011). In the formal setup, students lack experience in e-learning skills and abilities ( Poulová et al., 2011). Research studies have revealed that undergraduate students need extensive training in digital technologies ( Kim et al., 2018). This training is essential when students enter a blended learning environment, primarily pointing to the post-COVID education scenario. To moderate the gap, in institutions of higher learning, both learners and educators need to develop technology-related knowledge, skills, and attitudes through ongoing learning programmes ( Kim et al., 2018). Only agile ("agile" as used in the domain of technology) methodology and development referring to iterative processes and continuous improvement by building a culture of constant growth ( Himmelsbach et al., 2019) seems to be the viable solution. Students must embrace an agile mindset to meet the demands of digital innovations.

LA is an essential factor that integrates digital technologies into student learning and engagement in academic life. The theory of Learning Agility emphasizes that “individuals who have performed well in the past will not necessarily perform well in the future in a new job” ( Connolly, 2001). It is believed to significantly influence learners’ ability to progress to more complex and challenging learning assignments ( Almeida, 2019). Similarly, it can be presumed that students living in an era of transition may find it challenging to adapt to new learning situations with the present DC levels. Therefore, they are anticipated to be flexible and fast learners amid a high level of knowledge uncertainty posed by COVID-19 and evolving digitalization as prerequisites to seize new opportunities. The construct LA is more appropriate for consideration in this research context as its basis is rooted in adult learning and self-regulated learning ( Allen, 2016). Students perceive that agile practices have a great potential to enhance their learning experiences ( Melnik & Maurer, 2002). The definition of perceived learning, i.e. "changes in the learner’s perceptions of skill and knowledge levels before and after the learning experience", as given by Alavi et al. (2002), is appropriate in this context to ensure the quality of learning and improvement in the learning experience. Hence, as a predictor of students’ enriched learning experience, we hypothesize that LA mediates the relationship between DC and PL.

H2: There is a significant positive relationship between students’ Digital competence and learning agility in an online learning environment.

H3: There is a significant positive relationship between students’ learning agility and perceived learning in an online learning environment.

H4: The learning agility of students mediate the relationship between students’ Digital competence perceived learning in an online learning environment.

Based on the above literature, the following model ( Figure 1) is proposed.

Ethical approval was obtained from the Institutional Research and Ethical Committee of Welcomgroup Graduate School of Hotel administration (WGSHA), Manipal Academy of Higher Education via Reference No. WGSHA–IRC-2021-02 dated 14-08-2021. The committee waived the written consent since there was no risk involved for the participants, and most participants were above 18 years of age. Parental consent was also waived for a few participants of 17 years because of the no-risk nature of the study, and these underage participants were in the same cohort as the other participants, i.e., university students. Additionally, one of the authors visited the classrooms to explain the objectives and informed the participants that participation in the survey is voluntary. Thus, verbal consent was obtained before distributing the online survey form.

Data was collected from 359 full-time students across professional disciplines of a well-known private university in India. This university offers higher education in Medical, Paramedical, Allied Health, Health Science, Pure Science, Technology, Management, Hospitality Management, Commerce, Media, Humanities, Geopolitics, and few other disciplines. The diversity in the background was considered adequate to represent the different proficiency levels in DC among the student community. The questionnaire was developed in Microsoft Forms, and the web link of the online questionnaire was emailed to 1,200 students with an explanation on the constructs as well as study objectives. Though online, the researchers circulated the questionnaire to the students in the classroom. The research team members were present to explain the constructs of the study variables. The majority of the students filled out and submitted the questionnaire in the presence of the researchers. The data was collected in May 2021 and August 2021. A week after this, a follow-up email was sent as a reminder to expedite the data collection process. The data was collected in the month of May 2021 and August 2021.

In this cross-sectional research, the respondents were selected based on purposive sampling. The respondents have attended a minimum of 12 months of online classes. In total, 359 valid responses were received yielding a response rate of 30%. The sample included among the respondents, 224 (62.4 %) male and 135 (37.6%) female students. Among the respondents, 315 (87.7%) were undergraduates, and 44 (12.3%) were postgraduates.

The measuring instrument was developed after an in-depth literature review. The DC survey instrument was borrowed from ( Ferrari et al., 2013b). DigComp 2.1 provides a set of proficiency levels ranging from basic to advanced and is intended to provide a common understanding of what individuals should be able to do at each level. For example, at the basic proficiency level, individuals should be able to perform simple tasks such as sending and receiving emails, using search engines to find information, and creating simple documents. At the advanced proficiency level, individuals should be able to perform more complex tasks such as programming, data analysis, and creating interactive digital content. The proficiency levels are intended to provide a common understanding of what individuals should be able to do at each level and to help educators and trainers design learning experiences that are appropriate for the level of the learners. The 21 items were measured on a 5-point Likert scale where 1 represents “very low”, and 5 represents “very high”. A higher value would indicate a higher level of DC. Though the original DC framework is based on three levels, we have adopted a 5-point Likert scale ( Mehrvarz et al., 2021) to have uniformity in measuring all constructs. A higher value would indicate a higher level of DC. The LA (five items) was measured based on the scale of Kim et al. (2018). Respondents were requested to rate their agreement or disagreement with the statements on a 5-point Likert scale where 1 representing strongly disagree and 5 representing strongly agree. The outcome variable’s PL scale (six items) was adopted from the study by Narayan et al. (2021). These variables were operationalized using a 5-point Likert scale ranging from 1 (strongly disagree) and 5 (strongly agree). The respondents’ demographic details such as age, gender, and education were also included in the survey instrument. The full questionnaire can be found in the Extended data ( Mallya & Patwardhan, 2022b).

The Kaiser-Meyer-Olkin (KMO) test was used to test the sample adequacy. The KMO value is above the recommended value of 0.6 (0.93), and Bartlett’s test of sphericity is significant (χ ² (210) = 4478, p < .001), thus confirming the suitability of data for factor analysis ( Kline, 1994).

Reliability and validity (together known as psychometric properties) of the constructs (or factors) are the two prerequisite features in evaluating the measurement scale. This ensures the integrity and quality of a measurement scale. Before assessing the structural model, the first-order factor’s measurement model’s psychometric properties were assessed using the confirmatory factor approach. The model displayed good model fit indices (CFI = 0.95; TLI = 0.94; RMSEA = 0.05; SRMR = 0.05; x2/df = 2.64). The model was further tested for its reliability and convergent validity ( Table 1). Reliability was assessed based on the composite reliability (CR), and convergent validity was assessed based on the average variance extracted (AVE) values. According to Hair et al. (2014), the value of CR and AVE should be more than 0.70 and 0.50, respectively. All these values were above the recommended value ( Table 1), suggesting the constructs’ reliability and convergent validity. Further, except for the factor “Communication”, the model achieved discriminant validity ( Table 2). However, this is common due to the high correlation between the manifest indicators ( Koufteros et al., 2009; Marsh & Hocevar, 1985).

After achieving reliability and validity for the first-order factors model, the performance of the second-order factor model of DC was tested. Generally, first- and second-order CFA are conducted to validate the multi-dimensional scale. Specifically, when first-order factors act as indicators of second-order factors. Since DC comprises five sub-dimensions, the development of four models using a hierarchical approach was adopted to validate the second-order factor model ( Rindskopf & Rose, 1988). First, the single first-factor model with 21 items of DC was loaded (Model 1). The second model hypothesized that all the five dimensions of DC were separate and unrelated (Model 2). The third model (Model 3) hypothesized that all the five dimensions of DC were distinct but correlated. The fourth model (Model 4) was the second-order factor model of DC.

The hypotheses were tested using confirmatory factor analysis. The results are presented in Table 3. Table 3 shows that Model 1 and Model 2 did not have acceptable model fit indices. Further, Model 3 had marginally better model fit indices than model 4. Though model 3 had better fit indices, model 4, which hypothesizes a second-order factor model, was considered since it also had an acceptable fit.

The overall measurement model was tested using CFA after achieving desired model fit for the second-order factor. The purpose of the measurement model is to examine the relationship between the latent variables and their measures. The model indices values as per the recommended values (CFI = 0.94; TLI = 0.94; RMSEA = 0.04; SRMR = 0.05; x2/df = 2.37). The second-order factor model of DC was further tested for convergent and discriminant validity. The CR and AVE values were above 0.7 and 0.5, respectively ( Hair et al., 2014) ( Table 4). The discriminant validity of the constructs was tested by comparing the square root of AVE to bivariate correlation values between the constructs ( Table 5). According to ( Fornell & Larcker, 1981) square root of all measuring constructs should be greater than the bivariate correlation values between the constructs. The overall measurement model achieved discriminant validity.

After establishing the reliability and validity of the measurement model, the model fit indices of the structural model were tested ( Table 6). The purpose of the structural model is to test the proposed hypotheses in the study. The fit indices were within acceptable range (CFI = 0.928; TLI = 0.922; RMSEA = 0.0544; SRMR = 0.0604; x2/df = 2.04).

The structural model assessment was used to test the hypothesized relationship as conceptualized in the proposed model. This included the relationship between DC, LA, and PL. The R ² values (the coefficient of determination) and beta values (path coefficients) were the parameters used to determine the strength and magnitude of the relationship between the constructs. All path relationships were statistically significant ( Figure 2).

Hypothesis 1 (H1), proposing a significant positive relationship between DC and PL, was accepted (b = 0.33; p < 0.001), indicating that higher learners’ DC leads to higher learning outcomes. Similarly, hypothesis 2 (H2), which postulated the significant positive relationship between DC and learners’ agility also found support (b = 0.59; p < 0.001), suggesting that the higher DC of learners’ leads to a higher level of learning agility. The third hypothesis (H3) that proposed the positive relationship between the learners’ agility and PL also found support (b = 0.21; p < 0.001). This significant positive path reveals that higher learners’ agility leads to higher student learning outcomes.

The mediating effect of learning agility between DC and PL was analyzed using PROCESS macro model 4 ( Hayes, 2018). We have used the bootstrap method with 5000 re-samples to test the indirect effect as the sample size was adequate ( Zhao et al., 2010). It is found that LA has a mediating effect between their DC and PL (H4MFE-SAT-SWL: β= 0.1238, 95%, CI [0.0381, 0.216]).

Figshare: Students’ Digital Competence and Perceived Learning: The mediating role of Learner Agility, https://doi.org/10.6084/m9.figshare.20423496.v3 ( Mallya & Patwardhan, 2022a).

This project contains the following underlying data: •

Data.xlsx (the data set consists of four constructs: digital competence, perceived learning, learners’ agility, and self-efficacy).

Figshare: Digital competency_questionnaire.docx, https://doi.org/10.6084/m9.figshare.20423364.v2 ( Mallya & Patwardhan, 2022b).

This project contains the following extended data: •

Digital competency_questionnaire.docx.

Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0).