Learning Management Systems (LMS) generate extensive interaction data that, when strategically combined with Learning Analytics (LA) and Learning Design Analytics (LDA), hold significant potential for institutional analytics. However, conventional LA analyses often focus exclusively on isolated course-level behaviors, thereby neglecting important pedagogical design contexts. Addressing this gap, our study systematically aligns fundamental LMSderived engagement indicators with three theoretically grounded constructs: (1) Massive vs. Distributed Learning, (2) Workload, and (3) Active Learning. We analyzed anonymized LMS interaction logs from courses of a Spanish brick-and-mortar university. Employing exploratory data analysis, theory-informed and expert-guided refinement, as well as rigorous feature engineering, we extracted a concise set of easily replicable indicators. Our primary contribution lies in proposing a structured, theoretically aligned approach for deriving meaningful, pedagogically contextualized insights from minimal LMS data, using only standard activity-level log fields commonly available at most institutions. This lightweight analytic approach not only facilitates broader institutional adoption but also supports targeted instructional interventions and informed course design improvements. Future research directions include validating the robustness of these indicators across multiple institutional contexts and exploring their predictive capabilities regarding key educational outcomes such as student satisfaction, engagement quality, and academic performance.
Analytics and Learning
Learning Analytics (LA) and Learning Design (LD) have grown as complementary approaches to
improving teaching and learning in higher education. Learning design is typically defined as the planned
sequence of learning activities, resources, and assessments that reflect an instructor’s pedagogical intent
for a course. Learning analytics, on the other hand, entails collecting and analyzing data about learners’
interactions, typically from Learning Management System (LMS) logs, in order to gain insight and
enhance learning processes. Early on, researchers recognized the significant potential synergy between
these fields. In fact, [
LMS platforms generate large volumes of interaction data, typically analyzed through Learning Analytics (LA) to extract behavioral patterns. However, analyses commonly focus on single courses and often overlook their interpretation in the frame of their Learning Design (LD), limiting their relevance to
CEUR Workshop
ISSN1613-0073
support institutional decision-making, for example, related to the pedagogical model of the institution
or the needs for teacher training [
This work examines student interaction logs from a university’s Moodle platform encompassing
85 courses and over 68,000 activity records. Such a cross-course dataset provides a rich basis for
institutional analytics, revealing patterns that single-course analyses might overlook [
This type of data may supply two complementary types of insights: Learning Design (LD) indicators and Learning Analytics (LA) indicators. LD indicators reflect instructors’ intended student engagement, particularly the nature and diversity of learning resources and tasks assigned to students. Such indicators provide information about teachers’ design choices and intended teaching methods. LA indicators, on the other hand, provide empirical insights into actual student engagement, such as LMS usage frequency, interaction patterns, inferred workload, and engagement patterns over time (e.g., distributed vs. massive). Analyzing these indicators together can help institutions better understand how course design choices align with, or difer from, actual student engagement behaviors and outcomes.
In this context, despite significant advances in Learning Analytics and its growing institutional importance, the practical challenge remains: how can institutions systematically utilize readily accessible LMS data to generate meaningful insights linked explicitly to educational designs and strategies? More specifically, there is a critical gap in translating basic, available activity-level LMS data into pedagogically relevant engagement metrics. Addressing this gap is essential for enabling institutional stakeholders to better understand and enhance student learning experiences. Thus, the research question guiding this paper is: How can basic activity-level LMS variables be systematically transformed into engagement indicators that map onto the constructs of Massive vs Distributed learning, Workload, and Active Learning?
To systematically address our research question—How can basic activity-level LMS variables be systematically transformed into engagement indicators that map onto the constructs of Massive vs Distributed learning, Workload, and Active Learning?—we adopted a methodological approach grounded in clearly defined extraction rules and modern analytical techniques. This section describes the LMS-derived dataset used in this study, outlines the theoretical constructs guiding our analysis, and specifies formal procedures and mathematical definitions developed to transform raw LMS log data into pedagogically meaningful indicators.
Interaction data collected by Learning Management Systems (LMS) is usually standardized and
available to system administrators across educational institutions. Such datasets are ideal for institutional
analytics due to their structured nature and the ease with which meaningful metrics can be extracted
and interpreted. The dataset used in this study is representative of standard LMS data of this type,
consisting of student interaction logs from Moodle aggregated at the Activity or Resource (A/R) level.
It contains six key variables described in Table 1. Although the original dataset contained information
regarding the time spent on each task, for our proposal, we decided not to consider it. Time spent on
tasks in an LMS might seem like a straightforward measure, but it often doesn’t tell the whole story.
Students might leave a page open while doing other activities, or spend more time on a task due to
dificulty without being engaged. This can lead to misleading conclusions about their involvement in
learning [
The selected indicators, as well as their classification into clear theoretical constructs, present a methodological reflection that connects Learning Design Analytics (LD) and Learning Analytics (LA). LD focuses mostly on instructors’ intentions and pedagogical choices, such as the variety and type of activities or resources ofered. In contrast, LA reflects actual student behaviors and interactions as captured by LMS data, such as engagement patterns, frequency of interaction, and distribution of work. By systematically aligning these empirically measured behaviors (LA indicators) with their corresponding pedagogical intentions (LD indicators), institutions can better assess the efectiveness and alignment of their instructional designs with actual student engagement outcomes. Analyzing discrepancies or congruencies between designed engagement (intended LD) and observed interaction patterns (empirical LA) allows for targeted instructional interventions and data-driven course design.
Following a basic statistical exploration of the dataset and theory-oriented alignment with ultimate
learning design intentions, a set of key analytical indicators was identified and defined to help
characterize student engagement [
Massive versus Distributed Learning Indicators aim to distinguish between highly concentrated patterns of interaction (massive learning) versus more distributed, sustained engagement patterns over time. Indicators in this category include the temporal spread of interactions with Activities/Resources (A/R Interaction Span), the frequency of visits per activity/resource (A/R Interaction Frequency), intervals between students’ first-time access to activities/resources ( A/R First-time Access Intervals), and overall course-level frequency of interactions (Course Interaction Frequency).
Workload Indicators were identified based on their potential link to student-reported satisfaction with course workload. These indicators include the breadth or variety of resources accessed, measured by the Diversity Activity/Resource Index (Diversity A/R Index)—higher values suggesting greater cognitive demand—and two frequency-based metrics: the total frequency of interactions per course (Course Interaction Frequency) and the average daily frequency of student access (Daily Access Frequency).
Active Learning Indicators are defined as measures reflective of student engagement behaviors that involve deliberate, frequent, and diverse interactions. Indicators considered here include the total number of unique activities/resources accessed (Diversity A/R Index), the total number of days students remained actively engaged with course materials (Course Interaction Span in days), regularity of intervals between initial student accesses to new activities/resources (First-time Access Interval), and the daily frequency of LMS access (Daily Access Frequency).
This section provides a standardized specification of the extraction rules required for computing the proposed engagement indicators from standard LMS logs. To ensure reproducibility, we formalize the raw data present in the event log for each activity, denoted as: = 〈 , , , ,
, 〉 ∶ = 1, ...., to the activity follows:
Where is the whole log for the activity a, represents an anonymized (or pseudoanonymized) student identifier, denotes the Activity/Resource identifier (A/R), is the number of visits for an event, and and are the dates (in days, YYYY-MM-DD format) of the first and the last student access , respectively. At the course level, the extraction rules for the indicators are defined as The Course Interaction Span (CIS) measures the total span in days of student engagement across the entire course and is calculated by subtracting the earliest first-access date last-access date. Formally, = (
) from the latest = 1 + ( − )
With dates computed as calendar-day diferences, inclusive of both endpoints. In parallel, Course Interaction Frequency (CIF) reflects overall engagement intensity as total visits of a student to any A/R during the course, divided by the course-interaction span, expressed as:
The Daily Access Frequency (DAF) quantifies the regularity of student engagement, calculated by counting the number of distinct calendar days on which at least one activity was accessed within the course. To compute this, we create a set of distinct dates of access per student per activity (from to ) and then count unique days across all activities in the course. To define this indicator formally, we need to define the set of days where the engagement happened:
Γ = { ∈ ℤ ∶ ⩽ ⩽ }
That is, every calendar day on which a student registered an interaction with any activity . Then, we need to aggregate across all records in course .
= ∑ Γ = ⋃=1 Γ
= |Γ | So, finally, we can define the daily access frequency for course as the cardinality of this union:
On the other hand, the Diversity A/R index (DAI) is a Learning Design indicator, specially designed to measure, to some extent, the intended breadth of usage of A/R from the perspective of the teacher. It could be calculated simply by the cardinality of the distinct A/R accessed within the course. Formally,
= |{ ∶ = 1, ..., }|
We can also aggregate some A/R indicators to the course level to allow a better alignment with other institutional insights. For instance the A/R-interaction span (AIS) quantifies the span of student engagement per A/R, but the average of all the accessed activities ∈ reports, for course , the average number of calendar days over which students engaged with each A/R, let’s denote by and the earliest first-access date and the latest last-access date, respectively, First compute the span of each activity as: = 1 + ( − )
= 1 | | ∑∈
= ∑{∶ = } =
( − ) = 1 | | ∑∈
Similarly, the A/R interaction frequency (AIF) follows the same aggregation process. Specifically, for every course , let denote the set of distinct activities/resources that registered at least one access, and let,
Be the total number of visits to the activity (Summing up the visit count over all rows i whose Activity ID equals a). With and as the earliest first-access and latest last-access dates for the activity , the visit rate for that activity is:
defined in calendar days (inclusive). The course-level A/R-interaction span (AIS) is then the mean of these activity spans.
The course-level A/R-interaction frequency (AIF) is then the mean of these per-activity rates: Therefore,
reports the average of the students’ visits to each activity/resource aggregated at the course level. Lastly, the First-time access interval (FAI) measures the regularity with which students engage for the first time with diferent activities/resources. First, we sequence all activities/resources in chronological order according to their earliest first-access date. Then we calculate the average interval (in days) between successive first-access dates =
1 | |−1 , +1
, +2 , etc. ∑ | |−1 =1 ( +1 − ) distinct activities to be interpretable.
Where activities are ordered in increasing
These extraction rules rely exclusively on standardized activity-level LMS data commonly available to institutions. They ofer straightforward implementation using standard computational tools (e.g., Python, R, SQL). Importantly, while the indicators proposed here are robust and coherent across typical LMS datasets, their interpretative clarity may vary slightly depending on the course structure (e.g., fully asynchronous or self-paced courses). Nonetheless, the described approach ofers a comprehensive and reproducible method for systematically assessing student engagement, facilitating integration between Learning Analytics and Learning Design Analytics at the institutional level.
order. Note that this indicator requires at least two
All the proposed indicators rely exclusively on fields commonly provided by standard LMS exports
(e.g., Moodle or Blackboard), namely student ID, course ID, activity ID, first and last access dates,
and total visit counts. No advanced data streams, such as xAPI or detailed server-session logs, are
necessary, making this approach particularly viable for institutions with limited LMS data availability.
The computational extraction of indicators can be eficiently carried out using standard tools such as
SQL queries (leveraging common functions like GROUP BY, MIN, MAX, and date diferences), as well
as Python’s Pandas library (groupby().agg()), or R’s dplyr package (summarise()); our 68,000-record
dataset required less than 10 seconds of processing time using each of these methods using a single
Apple Silicon M1 SoC. To ensure consistency, timestamps should always be converted to the institutional
time zone before computing spans, thus preventing inflated intervals due to mixed time zones. If
explicit visit counts are missing from LMS logs, assigning a default visit value of one per access event
maintains indicator stability; our sensitivity tests showed this simplification had minimal impact on
indicator rankings. Although the LMS collects a Time Spent ”Time on Task” field (HH:MM:SS per
activity), we deliberately exclude it from our indicator because raw dwell time recorded by LMSs is
widely acknowledged as an often unreliable standalone indicator of genuine, multidimensional student
engagement. [
The central aim of this study was to address the research question: How can basic activity-level
LMS variables be systematically transformed into engagement indicators that map onto
the constructs of Massive-versus-Distributed Learning, Workload, and Active Learning The
approach presented in Section 2 describes a practical, rigorously specified workflow that transforms
a minimal set of LMS log fields—first and last access dates, visit counts, and activity identifiers—into
interpretable engagement indicators. By expressly tying each metric to a pedagogical construct, the
paper illustrates, in a standardized and replicable manner, how design-aware analytics can be derived
without relying on complex or sensitive data. Temporal dispersion measures (Course-Interaction
Span, A/R-Interaction Span, First-time Access Interval) characterize massed versus distributed learning
behaviours, while frequency-based metrics (Course-Interaction Frequency, Daily-Access Frequency)
and breadth metrics (Diversity A/R Index) ofer proxies for workload demands and active-learning
opportunities. Together, these examples demonstrate the analytical power latent in even the most
rudimentary LMS exports and provide a concrete template that institutions can adopt or adapt to their
contexts [
However, despite the methodological feasibility, the broader adoption of these indicators faces
significant practical challenges. Higher Education Institutions (HEIs) are often cautious about sharing
ifne-grained LMS data with external or even internal analytics providers due to legitimate privacy
concerns, regulatory compliance issues such as GDPR, administrative overhead, and institutional risk
perceptions [
Additionally, our analysis underscores a fundamental limitation: LMS behavioral data alone are insufifcient to capture the complete complexity of the theoretical constructs we aim to measure. Indicators derived solely from LMS logs provide valuable but inherently partial views of student engagement, as they lack insights into underlying motivations, cognitive processes, and afective responses. For instance, indicators measuring the breadth and frequency of resource access (e.g., Diversity A/R Index) cannot independently distinguish between meaningful exploratory behavior and superficial resource navigation driven by confusion or lack of clarity. To address acknowledged limitations in current analytics practice, we propose expanding our minimal-data indicators into a multidimensional evidence set that combines (i) additional LA traces—such as assessment scores, formative-quiz attempts, and forum interaction networks, (ii) explicit LDA descriptors drawn from design documentation—planned weekly workload, task complexity, collaboration mode, and (iii) outcome variables—including course grades and validated satisfaction. This richer data integration directly addresses the shortcomings identified in recent reviews: Kaliisa et al. highlight how small samples, single-source logs, and nonstandardised outcomes reduce the evidential value of many dashboard studies [15], while Topali et al. show that instructor-facing dashboards frequently raise awareness but rarely translate into actionable pedagogical feedback because the displayed metrics lack a clear theoretical link to learning design [17]. Our approach answers this missing connection by explicitly mapping each indicator to the constructs of Massive-versus-Distributed Learning, Workload, and Active Learning, resulting in a stronger foundation for actionable, design-aware analysis.
In summary, the paper ofers three key contributions: (1) a rigorously defined, minimal-data workflow that converts standard LMS logs into theoretically grounded engagement indicators; (2) a discussion of practical considerations for institutional adoption, particularly around privacy and administrative feasibility; and (3) a call for multidimensional analytics that integrate additional Learning Analytics and Learning Design data to overcome the explanatory limits of single-source logs. Addressing these challenges will enhance higher-education institutions’ capacity to generate actionable, design-aware insights that ultimately improve teaching and learning outcomes.
This paper addressed the practical research question: How can a minimal set of activity-level LMS variables be systematically transformed into engagement indicators that map onto the constructs of Massive-versus-Distributed Learning, Workload, and Active Learning? To answer this, we developed a clearly defined and rigorously specified extraction workflow that converts only four standard LMS fields—first-access date, last-access date, visit counts, and activity identifiers—into three coherent indicator categories: temporal-dispersion, frequency, and breadth. By explicitly aligning each metric with established learning-design theory, this approach provides institutions with a standardized, replicable, and privacy-conscious analytic approach suitable for immediate implementation. Demonstrated feasibility across 85 university courses illustrates the practical potential of these indicators to yield meaningful insights, including identification of engagement patterns (massed versus distributed learning), workload peaks, and active learning opportunities, without requiring sensitive student data or advanced tracking capabilities.
Nevertheless, this study also highlighted important areas for future research. First, expanding towards a multidimensional analytics strategy is crucial: subsequent studies should integrate our minimal LMS indicators with complementary learning analytics signals (e.g., quiz results, forum interactions) and explicit learning-design metadata (planned weekly workload, collaborative task tags), providing richer interpretations and deeper insights. Second, validating the predictive power of these indicators by linking them to critical student-level outcomes such as performance (grades), satisfaction, and retention will further solidify their practical utility. Third, replicating the analysis across multiple institutions and diverse LMS platforms will test indicator robustness and generalizability. Fourth, although we excluded dwell-time measures due to validity concerns, future eforts could explore calibrated measures of engagement duration as supplementary indicators. Lastly, developing and trialing low-overhead dashboards that embed these indicators directly into instructor workflows will illuminate how designaware analytics influence pedagogical decisions and ultimately student learning outcomes. Taken together, these future directions will transform this initial proof-of-concept into a robust, scalable, and institutionally actionable analytics toolkit for enhancing higher education teaching and learning.
This manuscript includes content that was revised with the assistance of ChatGPT, a large language model developed by OpenAI. The model was used to assist with improving the clarity, conciseness, and style of the manuscript. All content was critically reviewed and validated by the authors to ensure accuracy and academic integrity. George Washington University, One Dupont Circle, Suite 630, Washington, DC 20036-1183, 1991.
Accessed: May 31, 2025. [Online]. Available: https://eric.ed.gov/?id=ED336049. [13] S. Freeman, et al., Active learning increases student performance in science, engineering, and mathematics, Proceedings of the National Academy of Sciences 111 (2014) 8410–8415. doi:10. 1073/pnas.1319030111. [14] M. Prince, Does active learning work? a review of the research, Journal of Engineering Education 93 (2004) 223–231. doi:10.1002/j.2168-9830.2004.tb00809.x. [15] R. Kaliisa, K. Misiejuk, S. López-Pernas, M. Khalil, M. Saqr, Have learning analytics dashboards lived up to the hype? a systematic review of impact on students’ achievement, motivation, participation and attitude, in: Proceedings of the 14th Learning Analytics and Knowledge Conference, ACM, Kyoto, Japan, 2024, pp. 295–304. doi:10.1145/3636555.3636884. [16] L.-A. Lim, A. Atif, K. Heggart, N. Sutton, In search of alignment between learning analytics and learning design: A multiple case study in a higher education institution, Education Sciences 13 (2023). doi:10.3390/educsci13111114. [17] P. Topali, I.-A. Chounta, A. Martínez-Monés, Y. Dimitriadis, Delving into instructor-led feedback interventions informed by learning analytics in massive open online courses, Journal of Computer Assisted Learning 39 (2023) 1039–1060. doi:10.1111/jcal.12799.