Higher education instructors deploy learning analytics and generative artificial intelligence in distinct cognitive and pedagogical configurations when designing academic courses, according to an empirical investigation conducted at a major Australian university. While higher education institutions can leverage the combined potential of learning analytics and generative artificial intelligence to enhance learning design practices, how both technologies are used in combination is not well understood. To investigate how university instructors actually reason through design choices under the affordances of these emerging tools, a research team applied quantitative ethnography to analyze eleven faculty focus groups. Interpreting instructional discourse through the theoretical architecture of Self-Determination Theory, the researchers examined how technical affordances intersect with educators’ fundamental psychological needs for autonomy, competence, and relatedness. The resulting models demonstrate that instructional teams do not treat predictive trace data and generative models interchangeably; instead, instructors isolate analytics for iterative course-level monitoring, contextual problem-solving, and learner support, whereas they channel generative artificial intelligence toward drafting localized assessment tasks, formulating evaluative rubrics, and structuring pedagogical opportunities for student self-determination.
The methodological framework developed by the researchers utilized quantitative ethnography to unite grounded qualitative thematic exploration with rigorous mathematical network modeling. These codes captured essential pedagogical dimensions: contextual information, technological roles, learning activity design, course-level design, assessment design, creative problem-solving, student support, decision-making autonomy, instructional design approach grounded in student self-determination, and instructor emotion. The coded discourse segments were subsequently modeled through Epistemic Network Analysis, an analytical technique engineered to quantify, visualize, and statistically test the co-occurrence patterns of cognitive codes within moving conversational units.
Statistical verification of the Epistemic Network Analysis demonstrated high structural fidelity, with co-registration correlations reaching 0.96 for Pearson and 0.96 for Spearman across the first singular value decomposition dimension, matched identically by 0.96 Pearson and 0.96 Spearman correlations across the second dimension. These elevated goodness-of-fit metrics confirmed that the low-dimensional projected space accurately represented the complex multidimensional connections embedded in the focus group discourse. Nonparametric testing via the Mann-Whitney U test established a statistically significant divergence between learning analytics discourse and generative artificial intelligence discourse along the first dimension, designated singular value decomposition dimension one (Mann-Whitney U = 370.00, p < 0.001, effect size r = -0.96). Along this primary horizontal axis, the median coordinate position for the learning analytics group rested at -0.23 across twenty-one units, whereas the generative artificial intelligence group demonstrated a median position of 0.32 across eighteen units. Conversely, the vertical axis, representing singular value decomposition dimension two, showed no statistically significant separation between groups (Mann-Whitney U = 200.00, p = 0.77, effect size r = 0.06), with the learning analytics median recorded at 0.04 and the artificial intelligence median recorded at -0.05.
When investigating how decision-making factors associated within discussions centered on learning analytics, the network structure revealed dense relational clusters anchored by educational context. The empirical network demonstrated that instructors’ most prominent conceptual pairings occurred between contextual information and technology role, which exhibited a connection strength edge weight of 0.32, contextual information and creative problem-solving at 0.32, and contextual information and course-level design at 0.23. Moderate edge weights characterized the intersections connecting technological role with learning activity design at 0.17, student support with technological role at 0.19, and decision-making autonomy with context at 0.14. Lower co-occurrence frequencies appeared between technological role and course-level design at 0.11, technological role and student self-determination design at 0.09, technological role and assessment redesign at 0.08, and technological role and course design decision-making autonomy at 0.07. The prominent triad among context, course architecture, and creative problem-solving highlights that instructors rely heavily on analytics dashboards to diagnose overarching curricular performance within their specific institutional environments.
Qualitative analysis supporting these quantitative edge weights clarified that instructors interpret trace data through the specific contours of their pedagogical goals rather than adopting algorithmic readouts uncritically. Faculty reported examining visual performance dashboards to identify learning friction, disengagement patterns, or difficult concepts where cohorts required additional academic scaffolding. However, the data confirmed that educators experienced their decision-making autonomy as heavily bounded by institutional mandates and administrative constraints. Negotiating these systemic pressures required collaborative problem-solving, affirming theoretical perspectives on complexity leadership where shared autonomy and clear contextual communication enable teaching professionals to resolve instructional obstacles without losing programmatic coherence.
In contrast to the macroscopic, monitoring orientation observed in analytics discourse, instructors’ deliberations surrounding generative artificial intelligence coalesced tightly around granular assessment construction and pedagogical self-determination. The Epistemic Network Analysis for artificial intelligence identified its strongest connections between contextual information and creative problem-solving at an edge weight of 0.26, and between technology role and assessment design, also registering an edge weight of 0.26. The association between technological role and design for student self-determination formed another prominent link at 0.15. Secondary associations linked learning activity design with technological role at 0.15, and technological role with course-level design at 0.14. Notably, the network demonstrated marginal associations between decision-making power and technological implementation at 0.06, technological role and student support at 0.03, and decision-making power and student support at 0.01. This sparse connectivity confirms that university instructors rarely conceptualize generative models as immediate mechanisms for pastoral student care or institutional governance.
Instructors reported employing generative artificial intelligence to create first drafts of assessment elements, specifically case studies and rubrics. This practical application mirrored principles of Self-Determination Theory, as faculty engaged in collaborative co-design processes with students, particularly within pre-service teacher education cohorts. In these environments, instructors engaged future teachers in authoring rubrics and refining generative outputs, thereby directly supporting student autonomy and instructional competence. Faculty avoided employing generative systems to generate top-down course frameworks or core learning objectives, expressing greater confidence in human expertise for structural constructive alignment while utilizing artificial intelligence for localized ideation. Furthermore, the educators’ sense of design agency remained shaped far more by immediate departmental culture and operational guidelines than by the capabilities of generative technology itself.
A formal subtraction network analysis directly compared the topological networks of learning analytics and generative artificial intelligence, isolating the defining conceptual divergences between the two computational domains. The resulting subtractive model revealed that learning analytics generated substantially stronger associations between technological role and student support, yielding a differential edge weight of 0.16. Analytics discussions also maintained stronger ties between technology role and course structure at -0.03 in differential weight, activity design and technology at -0.02, and decision-making autonomy and technology at 0.01. Conversely, the generative artificial intelligence discourse exhibited pronounced differential dominance regarding assessment design, with the subtraction weight between technology and assessment reaching -0.18, alongside a -0.06 difference for student self-determination design. Although creative problem-solving remained central to both technologies, its co-occurrence with technological implementation was more intense in artificial intelligence discussions, registering a differential value of -0.13, while contextual considerations retained an indispensable organizing presence across both technological spheres.
The study also traced the presence of instructor emotional expressions across both cohorts, uncovering distinct affective dynamics during technological adoption. Within learning analytics discussions, emotional expressions maintained low but consistent ties with decision-making autonomy at 0.01, learning activity design at 0.02, and student support at 0.02. In generative artificial intelligence sessions, emotional expressions yielded slightly more diverse connections, linking to assessment design at -0.04, technological role at -0.03, creative problem-solving at -0.03, and student self-determination design at -0.01. The focus groups revealed that discussions of artificial intelligence evoked a noticeably heightened emotional response from teaching staff compared to conventional data dashboards. From the perspective of Self-Determination Theory, giving instructors the opportunity to explore emotional responses without judgment or control could lead to better integrative emotional regulation, resulting in enhanced decision-making and creativity.
These empirical findings yield direct institutional implications for university executives, educational developers, and academic leaders seeking to synthesize digital technologies into coherent pedagogical policy. The researchers emphasize that higher education institutions should implement distributed leadership models that share decision-making power between administrative bodies and frontline teaching staff regarding software procurement, curricular governance, and academic integrity guidelines. Enabling transparent institutional communication networks allows effective design strategies to circulate freely rather than remaining isolated within individual faculties. When assessing course-level analytics, instructors may identify patterns of student difficulty and subsequently leverage generative artificial intelligence to help generate targeted support material.
The investigators explicitly outlined several methodological limitations that contextualize their empirical conclusions and define future research trajectories. Participant recruitment was restricted to a single faculty consisting of instructors teaching within pre-service and enabling programs at one Australian institution, with fourteen individuals participating across both analytics and artificial intelligence focus groups. The context of pre-service teaching influenced results regarding participants engaging in co-design with their students, which may not necessarily transfer to other discipline contexts. Additionally, initial pre-moderation Cohen’s kappa values for decision-making autonomy and instructor emotion were low despite high raw agreement, an artifact stemming from the low natural base rate of these specific codes within conversational segments. Furthermore, the binary aggregation moving-window model employed in the Epistemic Network Analysis measured the presence of conceptual connections rather than their internal frequency or discursive intensity, and the findings represent inferential patterns derived from discourse rather than continuous direct observations of classroom execution.
Subject of Research: Higher education instructors' decision-making and learning design practices when integrating learning analytics and generative artificial intelligence
Article Title: Understanding the role of learning analytics and generative artificial intelligence on decision-making and learning design practice in higher education
Article References: Claassen, A., Ebbert, D., Kovanović, V., Mirriahi, N., & Dawson, S. (2026). Understanding the role of learning analytics and generative artificial intelligence on decision-making and learning design practice in higher education. International Journal of Educational Technology in Higher Education, 23(1), Article 43. https://doi.org/10.1186/s41239-026-00619-4
Image Credits: AI Generated
DOI: 10.1186/s41239-026-00619-4
Keywords: learning analytics, generative artificial intelligence, learning design, epistemic network analysis, quantitative ethnography, self-determination theory, higher education, assessment design, student support, course level design, creative problem solving, decision-making autonomy, instructor emotion, educational technology, pre-service teacher education, collaborative learning design
Cite Scienmag News
Wayne Ward. (November 18, 2026). Study Shows Instructors Use Learning Analytics for Support and Generative AI for Assessment. Scienmag. https://scienmag.com/study-shows-instructors-use-learning-analytics-for-support-and-generative-ai-for-assessment/
Wayne Ward. "Study Shows Instructors Use Learning Analytics for Support and Generative AI for Assessment." Scienmag, 18 November 2026, https://scienmag.com/study-shows-instructors-use-learning-analytics-for-support-and-generative-ai-for-assessment/. Accessed 18 November 2026.
Wayne Ward. "Study Shows Instructors Use Learning Analytics for Support and Generative AI for Assessment." Scienmag. November 18, 2026. https://scienmag.com/study-shows-instructors-use-learning-analytics-for-support-and-generative-ai-for-assessment/








