Sustainable Lifelong Learning Competence: Understanding University Students’ Self-Regulated Learning in Flipped Classrooms by Combining Questionnaire and Learning Analytics Data

Abstract

As an important sustainable lifelong learning competence, self-regulated learning involves a continuous process of self-monitoring and self-directing towards a learning goal. This study examined the level of alignment between university students’ self-regulated learning (SRL) profiles using questionnaire data and learning analytics data in flipped classrooms. On the one hand, a hierarchical cluster analysis using the questionnaire data generated two learning profiles of high and low self-regulated (SR) learners. On the other hand, a hierarchical cluster analysis using the questionnaire data produced two learning profiles of active and passive online learners. Although a cross-tabulation analysis showed a significant and positive relationship between students’ learning profiles identified using the questionnaire and learning analytics data, the association was rather weak. Of the high SR learners, there was a significantly higher proportion of active online learners than passive online learners. In contrast, among low SR learners, a significantly lower proportion of active online learners than passive online learners was found. Furthermore, high SR and active online learners and high SR and passive online learners had significantly better academic achievement than low SR and active online learners and low SR and passive online learners, demonstrating the importance of SR in flipped classrooms.

1. Introduction

UNESCO’s Education for Sustainable Development (ESD) initiative encourages individuals and institutions to critically rethink their perspectives and actions in order to contribute to a more sustainable future. A core objective of ESD is to expand access to high-quality education across all levels and sectors of society. By reshaping educational practices, ESD seeks to equip learners with the knowledge, skills, values, and attitudes necessary for sustainable living. In alignment with these goals, higher education institutions are increasingly reimagining and innovating teaching and learning through the integration of technological advancements, creating learning environments that are more adaptable, student-centered, and experiential. In this era of rapid technological change, many traditional, lecture-based courses have been transformed into flipped classrooms [1]. As a distinctive form of blended learning, flipped classrooms use in-class time primarily to deepen understanding, clarify key concepts, and contextualize knowledge through interactive, application-focused activities [2].

Flipped classroom models have become increasingly prevalent in contemporary higher education course design, as they not only promote active learning but also offer a range of pedagogical benefits [3]. One of their key advantages is flexibility: students can access online learning materials anytime and anywhere, enabling the approach to accommodate diverse learning styles and needs [4]. Moreover, flipped classrooms foster the development of essential graduate attributes, including problem-solving abilities, interpersonal communication, teamwork, and collaboration skills [5]. Compared with traditional instructional methods, flipped classroom designs have also been shown to improve student attendance, increase retention, enhance engagement in learning activities, and lead to better learning outcomes [6].

However, flipped classrooms do not always result in high levels of student satisfaction or desirable learning outcomes [7]. Achieving effective outcomes requires students to assume significant responsibility for and actively regulate their own learning, particularly in the online components of the course. They are expected to engage meaningfully with preparatory activities before attending face-to-face sessions. For instance, effective self-regulated learning (SRL) in the pre-class phase involves promptly reading assigned materials, watching recorded video lectures, completing self-assessments of the content, applying appropriate online learning strategies, managing time efficiently, and avoiding last-minute studying [8]. Therefore, developing a deeper understanding of students’ SRL processes in flipped classrooms can enable educators to provide structured guidance and scaffolding, ultimately enhancing students’ learning experiences and outcomes in such learning environments.

Compared with traditional course designs, students’ learning experiences in flipped classrooms are more complex, as they are required to navigate between face-to-face and online learning environments. Within these experiences, not only do students interact with humans, such as lecturers, tutors, and fellow students, but also engage extensively with online learning management systems, where they encounter a range of technology-enhanced learning tools (the material elements) [9]. Capturing the multifaceted nature of students’ learning in courses designed as flipped classroom experiences therefore necessitates collecting data from multiple sources, integrating both self-reported measures (e.g., questionnaires, interviews, and diaries) and learning analytics data [10]. Such a combined approach is particularly valuable for investigating students’ self-regulated learning (SRL) profiles, as it mitigates the limitations associated with the subjectivity of self-reported instruments, such as the Motivated Strategies for Learning Questionnaire (MSLQ; [11,12]), while complementing learning analytics with deeper insights into students’ internal states, including their motivation, self-efficacy, and anxiety [13]. Accordingly, the present study adopts a mixed-methods approach, integrating questionnaire data with learning analytics to develop a more comprehensive understanding of students’ SRL profiles in flipped classroom contexts.

2. Literature Review

2.1. Social–Cognitive Perspective of SRL

Self-regulated learning (SRL) is widely regarded as a crucial, sustainable lifelong learning competence. It is defined as “an active, constructive process whereby learners set goals for their learning and then attempt to monitor, regulate, and control their cognition, intentions, and behavior, guided and constrained by their goals and the contextual features of the environment” [14]. Drawing on multiple theoretical perspectives, scholars have examined a range of individual and contextual factors that influence how SRL unfolds [15]. Although definitions and interpretations of SRL vary, all major models share four core principles [14]. First, learners actively construct meaning, influenced by their prior knowledge and the context of learning. Second, they are capable of monitoring and adjusting their cognition and motivation. Third, they can strategically modify these processes to achieve their learning goals. Finally, learning emerges from the dynamic interaction between individual characteristics and environmental conditions. While all models emphasize learner autonomy and self-direction, they differ in the relative emphasis placed on specific components of the self-regulatory process [16].

Among various conceptualizations of SRL, the social–cognitive perspective is among the most widely recognized in educational research [17]. Building on social cognitive theory and Bandura’s triadic model of human functioning, Zimmerman proposed a triadic cyclical model of SRL, consisting of three interrelated phases: forethought, performance, and self-reflection [18]. Within this framework, SRL highlights learners’ proactive engagement in their own learning and the regulation of their motivation, emphasizing the interconnections among goal-setting, motivational beliefs, feedback, and self-efficacy. It also illustrates how self-efficacy and strategy use evolve cyclically across the three phases [19]. This model positions SRL as a multifaceted, dynamic process, underscoring the importance of metacognitive awareness, motivational engagement, and behavioral strategies in fostering effective self-regulation [20,21].

Because SRL is inherently dynamic and context-dependent, learners’ motivation and the strategies they employ can vary according to the learning context, the nature of the task, and their internal psychological states [22,23]. In recent years, growing scholarly attention has been devoted to examining SRL within diverse instructional contexts, particularly in online and blended learning environments [24,25].

2.2. Research Combining Questionnaire and Learning Analytics Data to Profile SRL in Online and Blended Courses

Recognizing the limitations of relying solely on either questionnaire data or learning analytics to understand students’ SRL profiles in online or blended learning environments, an increasing number of studies have adopted mixed-methods approaches that combine both data sources. This integration not only provides a more comprehensive and holistic understanding of SRL but also facilitates the triangulation and validation of findings [26,27].

For example, Li et al. analyzed learning analytics related to students’ time management and effort regulation behaviors, complemented by pre- and post-course surveys measuring self-reported time management and effort regulation [28]. Their results revealed significant correlations between students’ reported behaviors and their actual behaviors in the post-course survey, but not in the pre-course survey, indicating a degree of alignment between students’ self-perceptions and their observed strategies as the course progressed.

Similarly, Ye and Pennisi investigated the consistency between self-reported data (from surveys and interviews) and learning analytics data (tracking the frequency of students’ interactions with course content, peers, and instructors) in an asynchronous online agriculture course [27]. Their findings showed only partial consistency between the SRL profiles derived from the two data sources, with the overlap ranging from 30.8% (three-cluster solution) to 53.8% (two-cluster solution). Interestingly, some negative relationships also emerged. For instance, students who reported high SRL and frequent help-seeking behavior tended to complete fewer lectures and spent less time engaging with supplementary resources. Likewise, those who rated themselves highly on self-evaluation often interacted with fewer topics and demonstrated lower lecture completion rates.

In the context of blended course designs, Han et al. combined data from the Motivated Strategies for Learning Questionnaire (MSLQ) with learning analytics to examine Czech undergraduates’ SRL profiles [13]. The questionnaire data identified two profiles (i.e., students with stronger SRL and those with weaker SRL), while the learning analytics data revealed two distinct groups based on levels of online engagement (high versus low). Cross-tabulation analyses showed a strong association between the two approaches: most students identified as stronger self-regulators based on questionnaire data were also highly engaged online, whereas those with weaker SRL profiles tended to exhibit lower levels of online engagement.

2.3. Research Using Learning Analytics Data to Profile SRL in Flipped Classrooms

There is still limited research that combines questionnaire data with learning analytics to investigate students’ SRL profiles in flipped classroom settings. One notable study addressing this gap was conducted by Jovanović et al. [29], who used learning analytics to profile SRL strategies among 290 computer science students. The researchers identified five distinct SRL strategies, which differed not only in the types of learning activities students engaged in but also in the number of learning sequences, approximating the duration and intensity of their online learning:

• The intensive strategy was characterized by engagement in a wide variety of online learning activities and the largest number of learning sequences.
• The strategic strategy focused primarily on summative and formative assessment tasks and produced the second most learning sequences.
• The highly strategic strategy emphasized summative assessment tasks and reading activities and generated the third-highest number of learning sequences.
• The selective strategy concentrated mainly on summative assessment tasks with limited reading activities and produced the second-lowest number of learning sequences.
• The highly selective strategy involved engagement only in summative assessment tasks and resulted in the fewest learning sequences.

Jovanović et al. further examined students’ academic performance in relation to these SRL strategies [29]. Their findings revealed that students adopting the intensive, strategic, or highly strategic strategies achieved significantly higher scores on both midterm and final examinations than their fellow students who adopted the selective or highly selective strategies.

Similarly, Fincham et al. analyzed learning analytics data from more than 1000 students enrolled in flipped classrooms over three consecutive years [30]. Their findings revealed two distinct sets of SRL strategy profiles emerging at different stages of the course. In the first half of the semester, four SRL strategies were identified: a diverse strategy, a highly active strategy, a formative-content-oriented strategy, and a disengaged strategy. In the second half, students’ SRL strategies shifted to include a highly content-oriented strategy, an intensive strategy, an assessment-oriented strategy, and a highly assessment-oriented strategy. Notably, SRL strategies identified in the second half of the course were stronger predictors of students’ final exam performance than those observed in the first half.

Although the above studies provided valuable insights into students’ use of SRL strategies in flipped classrooms, these studies did not capture information about learners’ internal states, such as self-efficacy, motivation, and anxiety, which are essential components of SRL from a social–cognitive perspective. Addressing this gap, the present study combines questionnaire data with learning analytics to generate comprehensive SRL profiles and to examine how profiles derived from these two data sources differ. Specifically, the study aims to answer the following research questions:

(1)

How do students’ SRL profiles generated using questionnaire data differ from their profiles generated using learning analytics data?

(2)

How does students’ academic achievement differ based on their SRL profiles generated using questionnaire and learning analytics data?

3. Method

3.1. Sample and Research Context

The sample comprised 139 first-year university students (107 males and 32 females) enrolled in a computing engineering degree program at a public Australian university. All participants were taking an introductory course on computer systems delivered through a flipped classroom format.

The course incorporated both face-to-face and online components. The face-to-face component consisted of a two-hour weekly lecture, a two-hour weekly tutorial, and a three-hour weekly laboratory session. Lectures were designed to clarify complex theoretical concepts and demonstrate how these theories can be applied to solve practical problems. Tutorials provided students with opportunities to engage in problem-solving exercises and analyze mini case studies, while laboratory sessions focused on developing practical skills such as configuring computer systems and designing digital circuits. Previous research employing combined approaches to investigate students’ SRL has predominantly been conducted in either fully online or blended learning contexts, underscoring the relevance of examining SRL within a flipped classroom design.

The online component required students to participate in compulsory learning activities both before and after each week’s lecture, tutorial, and laboratory session, serving as preparation for and consolidation of the in-person learning. Five types of online learning activities were integrated into the learning management system (LMS):

• Pre-lecture readings: A combination of required and supplementary reading materials.
• Pre-lecture videos: Short video clips of pre-recorded lectures and demonstrations of problem-solving tasks to be covered in the upcoming tutorials and laboratory sessions.
• Pre-lecture quizzes: Assessments designed to evaluate students’ understanding of key theoretical concepts prior to class.
• Post-lecture resources: Web links, lecture notes, summaries of complex concepts, and detailed instructions for tutorial exercises and laboratory work.
• Post-lecture problem-solving tasks: Activities assessing students’ ability to apply theoretical knowledge to practical problems.
• Dashboard: An analytics tool that visualizes students’ progress and patterns of online learning, as well as their performance relative to the class average.

3.2. Data and Instruments

3.2.1. Questionnaire Data Collected Using the Motivated Strategies for Learning Questionnaire (MSLQ)

Five scales from the MSLQ [11,12] were used to measure students’ SRL. They were self-efficacy (8 items), intrinsic motivation (5 items), anxiety (4 items), metacognitive learning strategies (4 items), and cognitive learning strategies (4 items). The items were on 7-point scales with 1 indicating “never true of me” and 7 representing “always true of me”. A confirmatory factor analysis (CFA) was conducted to examine the factor structure of the MSLQ. The goodness-of-fit statistics were used as primary indices for the CFA due to the sensitivity of chi-square statistics to sample size. Specifically, the criteria of goodness-of-fit statistics for an acceptable model fit are: the values of CFI and TLI being greater than 0.900, and the value of RMSEA being below 0.800 [31]. The results of CFA suggested that the five-factor latent structure of the MSLQ resulted in a proper fit: CFI = 0.915, TLI = 0.901, and RMSEA = 0.060. For the latent construct, coefficient H scale reliability is recommended as it is more robust than Cronbach’s alpha [32]. Coefficient H scale reliability takes into account the different amounts of information contributed by different items to the overall scale. The values of the coefficient H were self-efficacy (8 items, α = 0.911), intrinsic motivation (5 items, α = 0.873), anxiety (4 items, α = 0.884), metacognitive learning strategies (4 items, α = 0.690), and cognitive learning strategies (4 items, α = 0.726). The value of coefficient H for the metacognitive learning strategies scale approached 0.700, and the rest of the scales were all above 0.700, suggesting that the MSLQ was reasonably reliable.

3.2.2. Frequency and Duration of Online Learning Activities Measured Using Learning Analytics Data

The learning analytics captured students’ identification numbers, frequency of the six types of online learning activities (i.e., pre-lecture readings, pre-lecture videos, pre-lecture quizzes, post-lecture resources, post-lecture problem-solving tasks, and dashboard), and the total duration of online learning.

3.2.3. Academic Achievement

Students’ academic achievement in the course was evaluated using four assessments: tutorial and laboratory performance (25%), a research project (15%), the mid-term exam (20%), and the final exam (40%). Overall course grades were calculated by aggregating scores across these four components.

3.2.4. Data Collection Procedure

Data collection was carried out in full compliance with the ethical guidelines established by the Human Ethics Committee of the researcher’s university. Prior to participation, students were informed that their involvement in the study was entirely voluntary, and written consent was obtained for three aspects: completing the questionnaire, retrieving data on their online learning activities from the learning management system (LMS), and accessing their course grades. Participants were assured that all data would be used exclusively for research purposes and that their identities would remain anonymous throughout the study. The questionnaire was administered near the end of the course to enable students to reflect more comprehensively on their learning experiences in the flipped classroom environment.

3.2.5. Data Analysis Methods

To answer the first research question, we conducted two types of analyses. First, we performed a hierarchical cluster analysis with means (Ms) of the five scales in the MSLQ to produce profiles of students’ SRL experience using questionnaire data. To determine the appropriate number of clusters, the values of the Squared Euclidean Distance between clusters were used [33]. Then, using the cluster membership as an independent variable, we performed one-way ANOVAs on the frequencies of students’ online activities and online learning duration. Second, we performed another hierarchical cluster analysis with frequencies of students’ online activities and online learning duration to generate profiles of students’ SRL experience using learning analytics data. Similarly, we ran a series of one-way ANOVAs to examine whether the Ms of the five scales differed between the clusters generated by the learning analytics data.

To provide an answer to the second research question, we first grouped students according to their cluster membership generated using both the questionnaire and learning analytics data. We then performed one-way ANOVAs on students’ scores on the four assessments and their course grades using the group membership as a between-subjects variable.

4. Results

4.1. Descriptive Statistics

The descriptive statistics, including the minimum, maximum, M, and SD of all of the questionnaire and learning analytics data and students’ academic achievement, are presented in

Table 1. Descriptive statistics.

Variables	Minimum	Maximum	M	SD
Questionnaire data
Self-efficacy	2.000	7.000	4.799	0.979
Intrinsic motivation	2.800	7.000	5.524	0.953
Anxiety	1.000	7.000	3.642	1.362
Metacognitive learning strategies	2.250	7.000	4.577	1.063
Cognitive learning strategies	1.500	7.000	4.401	1.164
Learning analytics data
Frequency of pre-lecture readings	138.000	2492.000	817.650	446.952
Frequency of pre-lecture videos	3.000	2890.000	335.160	395.217
Frequency of pre-lecture quizzes	2.000	610.000	159.580	134.125
Frequency of post-lecture resources	59.000	1182.000	420.820	235.769
Frequency of post-lecture problem-solving tasks	2.000	1083.000	219.180	187.389
Frequency of dashboard	1.000	233.000	39.410	43.394
Duration of online learning	1.000	93.000	15.640	18.945

.

4.2. Results of Research Question 1—How Do Students’ SRL Profiles Generated Using Questionnaire Data Differ from Their Profiles Generated Using Learning Analytics Data?

The hierarchical cluster analysis produced a range of two-cluster to four-cluster solutions. The values of Squared Euclidean Distance revealed a relatively large increase in the value of a two-cluster solution (25.895) compared to the three-cluster (14.422) and four-cluster (11.053) solutions, suggesting that a two-cluster solution was more appropriate. Cluster 1 students (n = 64) had significantly higher ratings on self-efficacy (F (1, 137) = 71.215, p < 0.001, η² = 0.342), intrinsic motivation (F (1, 137) = 80.970, p < 0.001, η² = 0.371), and use of metacognitive (F (1, 137) = 85.948, p < 0.001, η² = 0.386) and cognitive (F (1, 137) = 12.331, p < 0.001, η² = 0.083) learning strategies than cluster 2 students (n = 75), but the two clusters did not differ in anxiety (F (1, 137) = 0.639, p = 0.426, η² = 0.005). Therefore, cluster 1 students were referred to as high SR learners, whereas cluster 2 students were labelled as passive SRL learners.

Based on the two profiles identified using the questionnaire data, students’ online learning patterns using the learning analytics data were compared using one-way ANOVAs and the results are presented in

Table 2. The results of the one-way ANOVAs based on SRL profiles generated using questionnaire data.

Variables	High SR Learners (n = 62)	Low SR Learners (n = 83)	F	P	η2
	M	M
Questionnaire data
Self-efficacy	5.416	4.272	71.251	<0.001	0.342
Intrinsic motivation	6.150	4.989	80.97	<0.001	0.371
Anxiety	3.742	3.557	0.639	0.426	0.005
Metacognitive learning strategies	5.289	3.970	85.948	<0.001	0.386
Cognitive learning strategies	4.762	4.093	12.331	<0.001	0.083
Learning analytics data
Frequency of pre-lecture readings	845.800	793.640	0.468	0.495	0.003
Frequency of pre-lecture videos	446.030	239.480	9.842	0.002	0.068
Frequency of pre-lecture quizzes	194.110	129.360	8.255	0.005	0.058
Frequency of post-lecture resources	428.080	414.630	0.112	0.739	0.001
Frequency of post-lecture problem-solving tasks	274.580	172.770	10.672	0.001	0.074
Frequency of dashboard	48.560	31.070	4.562	0.035	0.041
Duration of online learning	21.000	10.780	7.992	0.006	0.073

. Table 2 shows that high SR learners significantly more frequently interacted with the four online learning activities than the low SR learners, including pre-lecture videos (F (1, 137) = 9.842, p = 0.002, η² = 0.068), pre-lecture quizzes (F (1, 137) = 8.255, p = 0.005, η² = 0.058), post-lecture problem-solving tasks (F (1, 137) = 10.672, p = 0.001, η² = 0.074), and dashboard (F (1, 137) = 4.562, p = 0.035, η² = 0.041). The high SR learners also spent significantly longer time studying online than low SR learners (F (1, 137) = 7.992, p = 0.006, η² = 0.073).

Based on the increased values of Squared Euclidean Distance, the hierarchical cluster analysis also retained the two clusters for the learning analytics data. The increase in the value of the two-cluster solution versus the three-cluster solution was 35.876, whereas the increase in the values of the three-cluster versus the four-cluster solution and the four-cluster versus the five-cluster solution were 15.868 and 12.580, respectively. Hence, the two-cluster solution was deemed to be more appropriate. Cluster 1 students (n = 74) significantly more frequently interacted with all of the online learning activities: pre-lecture reading (F (1, 137) = 87.547, p < 0.001, η² = 0.390), pre-lecture videos (F (1, 137) = 19.776, p < 0.001, η² = 0.129), pre-lecture quizzes (F (1, 137) = 60.051, p = 0.005, η² = 0.311), post-lecture resources (F (1, 137) = 73.757, p < 0.001, η² = 0.350), post-lecture problem-solving tasks (F (1, 137) = 77.798, p = 0.001, η² = 0.367), and dashboard (F (1, 137) = 25.587, p = 0.035, η² = 0.193). At the same time, cluster 1 students also spent significantly more time online (F (1, 137) = 30.222, p = 0.035, η² = 0.230) than cluster 2 students. According to the profiles of students’ online learning, cluster 1 was referred to as active online learners, whereas cluster 2 was designated as passive online learners. One-way ANOVAs were conducted to examine whether students’ SRL differed between active and passive online learners. The results in

Table 3. The results of the one-way ANOVAs based on SRL profiles generated using learning analytics data.

Variables	Active Online Learners (n = 74)	Passive Online Learners (n = 65)	F	p	η2
	M	M
Learning analytics data
Frequency of pre-lecture readings	1078.270	520.954	87.547	<0.001	0.390
Frequency of pre-lecture videos	466.343	183.159	19.776	<0.001	0.129
Frequency of pre-lecture quizzes	227.243	77.492	60.051	<0.001	0.311
Frequency of post-lecture resources	551.068	272.539	73.757	<0.001	0.350
Frequency of post-lecture problem-solving tasks	322.757	95.565	77.798	<0.001	0.367
Frequency of dashboard	54.147	14.976	25.587	<0.001	0.193
Duration of online learning	23.300	4.954	30.222	<0.001	0.230
Questionnaire data
Self-efficacy	4.966	4.608	4.771	0.031	0.034
Intrinsic motivation	5.727	5.292	7.548	0.007	0.052
Anxiety	3.659	3.623	0.024	0.878	0.000
Metacognitive learning strategies	4.885	4.227	14.581	<0.001	0.096
Cognitive learning strategies	4.389	4.415	0.018	0.893	0.000

show that active online learners reported significantly higher self-efficacy (F (1, 137) = 4.771, p = 0.031, η² = 0.034), higher intrinsic motivation (F (1, 137) = 7.548, p = 0.007, η² = 0.052), and used more metacognitive learning strategies (F (1, 137) = 14.581, p < 0.001, η² = 0.096) than passive online learners.

The results of the cross-tabulation analysis revealed a significant and weak association between the profiles generated using the questionnaire and learning analytics data: χ² (1) = 11.465, p < 0.001, φ = 0.287.

Table 4. Results of the cross-tabulation analysis.

Profiles generated using Questionnaire Data	Count % (Profiles by Analytics Data)	Profiles generated using Analytics Data
		Active Online Learners	Passive Online Learners	Total
High SR learners	Count % (profiles generated using analytic data)	44	20	64
		68.8%	31.2%	100.0%
Low SR learners	Count % (profiles using analytic data)	30	45	75
		40.0%	60.0%	100.0%
Total	Count % (profiles using analytic data)	74	65	139
		53.2%	46.8%	100.0%

shows that of 64 high SR learners, there was a significantly higher proportion of active online learners (68.8%) than passive online learners (31.2%). In contrast, among 75 low SR learners, a significantly lower proportion of active online learners (40.0%) than passive online learners (60.0%) was found.

4.3. Results of Research Question 2—How Does Students’ Academic Achievement Differ Based on Their SRL Profiles Generated Using Questionnaire and Learning Analytics Data?

Students were categorized into four groups: namely, group 1 = low SR and passive online learners (n = 45), group 2 = low SR and active online learners (n = 20), group 3 = high SR and passive online learners (n = 30), and group 4 = high SR and active online learners (n = 44). One-way ANOVAs and post hoc analyses were then conducted on the scores of the four assessments and the course grades using group membership as a between-subject variable.

The one-way ANOVAs showed that students’ scores on all four assessments differed significantly by group: performance on tutorials and laboratory practice: (F (3, 135) = 12.871, p < 0.001, η² = 0.222); research project: (F (1, 135) = 6.015, p < 0.001, η² = 0.118); and mid-term exam: (F (1, 135) = 8.028, p < 0.001, η² = 0.151); and final exam (F (1, 135) = 7.949, p < 0.001, η² = 0.150). There were also significant differences in students’ course grades by group (F (1, 135) = 13.596, p < 0.001, η² = 0.232). The results of post hoc analyses are presented in

Table 5. Post hoc analyses of students’ academic achievement by group.

	Tutorials and laboratory practice			p-values for pairwise comparisons
	n	M	SD	Group 2	Group 3	Group 4
Group 1	44	19.131	3.867
Group 2	30	16.604	6.035	0.056
Group 3	20	22.495	2.377	<0.001	<0.001
Group 4	45	21.299	2.676	0.031	<0.001	0.517
	Research project			p-values for pairwise comparisons
	n	M	SD	Group 2	Group 3	Group 4
Group 1	44	9.743	2.420
Group 2	30	9.240	3.204	0.887
Group 3	20	11.523	2.164	0.021	0.014
Group 4	45	11.361	2.709	0.020	0.015	0.994
	Mid-term exam			p-values for pairwise comparisons
	n	M	SD	Group 2	Group 3	Group 4
Group 1	44	12.840	3.176
Group 2	30	1.900	5.515	0.205
Group 3	20	15.100	2.833	0.050	<0.001
Group 4	45	15.020	3.638	0.030	<0.001	1.000
	Final exam			p-values for pairwise comparisons
	n	M	SD	Group 2	Group 3	Group 4
Group 1	44	17.890	6.368
Group 2	30	15.450	7.957	0.688
Group 3	20	24.230	8.982	0.007	0.002
Group 4	45	23.450	9.422	0.009	0.002	0.978
	Course grades			p-values for pairwise comparisons
	n	M	SD	Group 2	Group 3	Group 4
Group 1	44	59.734	11.794
Group 2	30	52.767	17.393	0.275
Group 3	20	73.801	13.379	<0.001	<0.001
Group 4	45	71.527	15.816	<0.001	<0.001	0.909

, which shows consistent patterns in the four assessments and course grades. While both group 4 (high SR and active online learners) and group 3 (high SR and passive online learners) had significantly higher scores on all four assessments and course grades than group 2 (low SR and active online learners) and group 1 (low SR and passive online learners); there were no significant differences between group 4 (high SR and active online learners) and group 3 (high SR and passive online learners) or between group 2 (low SR and active online learners) and group 1 (low SR and passive online learners).

5. Discussion

The aims of the present study were twofold: (1) to identify students’ SRL profiles in flipped classroom learning using both questionnaire and learning analytics data, and (2) to investigate how the profiles generated using the two types of data were consistent with one another. The results of our three types of analyses, namely, hierarchical cluster analysis, ANOVAs, and cross-tabulations, only showed a low level of consistency between the SRL profiles generated using the questionnaire and learning analytics data.

On the one hand, using the questionnaire data, the hierarchical cluster analysis identified two distinct profiles of high and low SR learners, who differed significantly in their reported efficacy, intrinsic motivation, and use of metacognitive and cognitive strategies in the flipped classroom course. Among either the high or the low SRL profiles, there was only a partial level of consistency between what learners reported about their SRL and what learners actually did in the online learning part of the flipped classroom course. The higher SR learners reported having higher efficacy, higher intrinsic motivation, and using more metacognitive and cognitive strategies in the course than lower SR learners. The observed online learning behaviors only showed a partial level of consistency with the self-reported data, as higher and lower SR learners did not differ in interactions with all of the online learning activities. Specifically, students with higher and lower SRL profiles differed in four out of six activities. Higher SR learners watched pre-lecture videos more frequently to prepare for the face-to-face learning and attempted pre-lecture quizzes more frequently to assess their understanding than lower SR students. The higher SR learners also attempted the post-lecture problem-solving tasks more actively to strengthen their understanding, monitored their online learning more frequently by checking the dashboard, and spent significantly more time studying online than their lower SR peers.

On the other hand, the hierarchical cluster analysis using the learning analytics data also detected two distinct active and passive online learning profiles in this flipped classroom course. While the active and passive online learners differed in the frequencies of interaction with all of the online learning activities as well as the total online learning duration, their reported levels of anxiety and use of cognitive strategies did not differ, which also indicated a partial level of consistency between the two types of data. The partial inconsistency between the SRL profiles identified using the questionnaire and learning analytics data was similar to the results obtained in Han et al.’s study with Czech students [13].

However, different from [13], which showed a strong association between the SRL profiles identified using the self-reported questionnaire and learning analytics data, only a weak association was found between the SRL profiles in our study. One possible explanation was that our learning analytics data consisted of indicators of both frequency and duration of students’ online learning, which presumably were more sensitive and accurate than the use of frequency information alone in the learning analytics data in [13]. Adding a time dimension to measure students’ online learning might enable the learning profiles to reflect some more nuanced features of students’ real learning behaviors than only using the frequency dimension [34,35].

Furthermore, we found that the percentages of overlap between the SRL profiles identified using the two types of data ranged from 60.0% to 68.8%, which were significantly lower than the percentages (ranging from 71.4% to 81.8%) in [13]. Our percentages, however, were closer to those found in Ye and Pennisi’s study (53.8%) [27]. Ye and Pennisi also used both frequency and duration aspects of students’ online learning behaviors to produce students’ SRL profiles [27]. Once again, these results seem to suggest that using a combination of frequency and duration types of the learning analytics data may more accurately reflect students’ actual learning behaviors than using either frequency or duration alone [35].

The partial consistency between the self-reported questionnaire data and observed learning analytics data supports the argument that the two types of data capture different aspects of students’ SRL experiences alone [13]. Indeed, previous studies have demonstrated that including both questionnaire and learning analytics data significantly explains more variance in students’ academic achievement in SRL than using a single type of data [27,28]. Therefore, to provide a more complete understanding of students’ SRL, it is recommended to collect both questionnaire and learning analytics data [36].

Corroborating with previous research, that higher SR learners tended to achieve better academic performance in flipped classroom learning than lower SR learners [8,29,30], our results also consistently demonstrated that higher SR learners not only scored significantly higher scores on all four assessment tasks, namely, tutorials and laboratory practice, research project, mid-term exam, and final exam, but also obtained significantly higher course grades than their lower SR peers. These results indicate the importance of SRL in formal education. Therefore, to help students succeed in formal education, institutions and educators should intentionally integrate developing students’ SRL through integrated curriculum [37] and intra-curriculum [38] support.

6. Limitations and Future Research Opportunities

Despite these intriguing findings, the study has several limitations that should be acknowledged and addressed in future research. First, neither the questionnaire data nor the learning analytics data used in this study captured changes in students’ SRL over the course. To better understand these dynamics, future studies could adopt a longitudinal design, which would allow researchers to explore how students’ SRL may evolve and fluctuate at the beginning, middle, and end of the semester and whether these fluctuations are associated with students’ academic achievement over time [28,30]. Another limitation of the study is the duration of online learning used in the study, which does not reflect how much time students spent on different online learning activities. Capturing duration in different online learning activities will help researchers and educators to obtain more nuanced features of students’ SRL in the flipped classroom, such as how much time students distribute between pre-lecture and post-lecture activities. Such information is valuable as previous research has consistently demonstrated the importance of preparation before attending face-to-face learning and teaching in flipped classrooms [1,34,39,40]. Last, future studies may also consider enhancing the reliability of the metacognitive learning strategies scale by trying to add additional items, as the coefficient H reliability of the metacognitive learning strategies scale was slightly low.

7. Conclusions

SRL is increasingly recognized as a vital lifelong skill that empowers individuals to take active control of their own learning processes [41]. It plays a crucial role in advancing sustainability in education by fostering learners’ capacity to take responsibility for their lifelong learning journeys. By developing SRL skills such as goal-setting, self-monitoring, and reflective thinking, learners become more autonomous and self-directed, enabling them to acquire new knowledge and skills as societal needs evolve [42]. In essence, SRL empowers individuals to become proactive, adaptive learners who can respond to emerging challenges, which is essential for building sustainable education systems and driving sustainable solutions [43]. In a rapidly changing world where knowledge and technologies evolve continuously, SRL ensures that individuals can adapt, reskill, and remain competent throughout their personal and professional lives.

To help students become SR lifelong learners, a prerequisite is to gain an understanding of students’ SRL profiles. This study investigated how well university students’ SRL profiles, derived from self-report questionnaires, aligned with those identified through learning analytics in flipped classroom settings. Considering our results, showing a relatively weak association between the two sets of SRL profiles generated using the two types of data, it is important to integrate both questionnaire-based measures and learning analytics data to gain a more comprehensive understanding of students’ SRL profiles.