glossaryPage.heroH1

Learning analytics is typically an institutional / administrative tool used by academic advisors, deans, and institutional research offices to identify patterns across student cohorts and target student-success interventions. Teacher-facing classroom analytics (gradebook reports, assignment-completion-rate per class, quiz performance distributions) is a separate category serving the teacher's in-class instructional decisions. The two overlap (a learning-analytics at-risk flag may trigger a teacher conversation with a student) but serve different decision-makers. Per Educause Learning Analytics Initiative, the institutional-administrator use case is the highest-confidence current deployment because the human-decision boundary stays clearly with advisors and student-success staff.

Significant. Per NEPC research and SoLAR (Society for Learning Analytics Research) peer-reviewed work, the major concerns are: (1) Demographic bias — predictive models trained on historical institutional data can encode historical bias against under-represented student groups, leading to disproportionate flagging that triggers institutional response. (2) Self-fulfilling prophecy — students flagged at-risk and treated as at-risk may internalise the label, with longitudinal effects on confidence and persistence. (3) Privacy — learning-analytics data is FERPA-protected; consent and disclosure rules apply to institutional use and to any data-sharing with third-party vendors. (4) Transparency — students often do not know that learning-analytics models are running on their data, which raises ethical questions about informed consent. Best practice per SoLAR guidance: publish institutional learning-analytics policy, retain human-in-the-loop intervention review, and audit per-demographic-group performance regularly.

Varies by institution and model maturity. Per Civitas Learning aggregate data and similar research, well-tuned predictive at-risk models identify roughly 60-80% of eventually-failing or eventually-withdrawing students 6-12 weeks before midterm grades, with false-positive rates of 10-20% (students flagged who would not actually have failed without intervention — though by design, the goal is to prevent failure so all flagged students get intervention). Per EAB research on early-alert systems, the model accuracy matters less than the institutional response loop — institutions with strong advisor-follow-up workflow see retention lift even with mediocre models; institutions with strong models but weak advisor response see minimal lift. The institutional capacity to act on the data matters more than the data itself.

SoLAR (Society for Learning Analytics Research) hosts the annual Learning Analytics and Knowledge (LAK) conference with peer-reviewed proceedings — primary research literature in the field. Educause Learning Analytics Initiative tracks US higher-ed adoption and publishes practitioner guidance. NEPC publishes critical-perspective research on bias and equity. AACE publishes peer-reviewed research on edtech ethics including learning analytics. The 2014 SoLAR-published "Ethics of Learning Analytics" framework remains a primary reference. Per UNESCO 2024 AI in Education guidance, learning analytics aligns with the administrator-facing AI category — high-confidence with appropriate governance.