Untested framework embeds ethics in university data decisions

A conceptual study introduces DDDM-FRAME, a six-layer model intended to organize data-driven decision-making in higher education while placing privacy, fairness, transparency, regulatory compliance and accountability across every stage of the process. Rather than treating ethics as a final check on an analytical product, the framework positions an Ethics and Governance Filter as a cross-cutting mechanism that accompanies data collection, integration, modeling, decision support, visualization and institutional action. The work synthesizes recent research on learning analytics, institutional analytics and responsible artificial intelligence, but it has not yet been tested in a real university, so its proposed benefits remain theoretical rather than demonstrated.

The need for such a structure follows from the growing use of student information systems, learning management platforms, data warehouses, dashboards and artificial-intelligence tools in university administration. These technologies can combine information on enrollment, attendance, academic performance, research activity, staffing and resource allocation. The study emphasizes that data-driven decision-making is broader than artificial intelligence: analytics may produce descriptive, predictive or prescriptive results, but institutional leaders must still interpret those outputs in light of managerial judgment, organizational priorities and local context. Analytical results are therefore treated as inputs to institutional decisions rather than decisions in themselves, with governance and managerial interpretation forming part of the process that translates evidence into action.

DDDM-FRAME was developed through a design science research approach, which is intended to produce a theoretically grounded solution artifact for a complex organizational problem. Its foundation came from a purpose-driven, thematic synthesis of literature published mainly during the previous decade, with earlier seminal work included where appropriate. Relevant studies addressed data-driven decision-making, learning analytics, institutional analytics, AI-supported decision support, responsible AI and governance in higher education. The review was not a systematic literature review and did not aggregate empirical findings; instead, it compared conceptual models and used recurring strengths, omissions and structural problems to guide the proposed architecture. Selection favored work with direct relevance to higher education institutions, a conceptual contribution to decision-making, governance or ethical dimensions, representation of analytical processes and potential applicability to institutional management. Studies focused exclusively on technical algorithm development without managerial relevance were excluded.

The comparative analysis examined six dimensions: ethical integration, governance structure, analytical capability, visualization support, institutional applicability and decision-support functionality. From this synthesis, the researcher identified three gaps: few frameworks addressed higher education administration holistically; ethical and governance considerations were not consistently integrated into analytical processes; and the relationship between analytical outputs and administrative practice remained unclear. DDDM-FRAME responds with an architecture that joins technical analysis to managerial decisions and treats ethics as an operational component of the workflow. Its development follows three stated principles: modularity, allowing layers to be evaluated or developed independently; holism, treating analytical, decision-support, visualization and ethical processes as interacting; and institutional applicability, allowing adaptation to different higher education contexts.

The framework’s first layer is the Data Acquisition Layer, where information may be drawn from student information systems, learning management systems and external repositories. The second, the Data Processing and Integration Layer, covers cleansing, merging and storage, including the resolution of missing values, inconsistencies and duplicate records. The third is the Analytical Intelligence Core, where descriptive, predictive and prescriptive methods identify patterns or estimate potential risks. The fourth, the Decision Support Engine, organizes analytical evidence into alternatives, potentially using methods such as decision trees or TOPSIS. These components do not automatically authorize action: the proposed design allows managers to consider institutional priorities and contextual factors, while the Ethics and Governance Filter applies throughout the sequence rather than waiting for a final managerial choice.

The fifth layer combines visualization and feedback through dashboards, reports and stakeholder response mechanisms. The sixth addresses outcomes and institutional impact, including implementation, organizational learning, performance monitoring and revisions informed by results. This creates a recursive process: institutional actions generate new information that can update data sources, models and decision support in later cycles. The sixth layer is intended to close the conceptual loop between a decision and its consequences, extending governance beyond algorithm assessment to the monitoring and revision of institutional action. Visualization is likewise treated as part of the decision cycle rather than merely a display stage, because stakeholders can examine the information, question its implications and contribute feedback for subsequent analysis.

Its principal distinguishing feature is the Ethics and Governance Filter, which operates across all six layers instead of appearing as a seventh isolated stage. At acquisition, it raises questions about which data are collected and how privacy is protected. During processing and integration, it considers data quality, consistency and missing values. At the analytical stage, it calls for assessment of potentially discriminatory model patterns and transparency; at decision support, it requires the basis of recommendations to be understandable and open to institutional challenge. Oversight also continues into visualization, implementation and outcome monitoring. The study suggests that such functions could involve data scientists, institutional managers, quality-assurance units and ethics committees, forming an interdisciplinary socio-technical oversight arrangement. This proposal specifies possible functions, but the paper does not present a validated procedure or an empirically tested implementation of the filter.

A hypothetical case illustrates the intended operation at a medium-sized public university facing increasing first-year dropout and performance disparities among programs over the preceding three years. In this example, the rectorate and quality-assurance unit collect academic results, attendance, online learning interactions and demographic information from institutional systems. These records are cleaned and combined in a data warehouse, after which illustrative models examine patterns associated with retention and program performance. Student risk profiles are modeled using prior achievement, attendance and learning interactions. A decision-support system could organize possible interventions, such as stronger first-year advising, targeted support or review of programs with persistent disparities. The authors stress that the risk estimates are illustrative applications, not empirically tested predictions or validated estimates of predictive accuracy, and that the decision-support engine organizes evidence without automatically determining institutional action.

In the same scenario, dashboards present dropout rates, risk patterns and trends for senior managers, while feedback from academic committees and other stakeholders allows institutional actors to question the information and its implications. Potential actions are then implemented, and their effects on retention, program performance and related indicators are monitored. Privacy safeguards, fairness analysis, bias review, transparency, regulatory compliance and accountability remain present throughout this sequence rather than being appended after a model has produced its result. The intended result is a traceable cycle in which institutional actors can ask not only how a recommendation was produced, but also whom it may serve, what consequences may follow and how the decision can be revised. This part of the paper is a demonstration of the artifact’s intended logic, not a trial showing that the framework produces better decisions.

The study presents DDDM-FRAME as a theoretical contribution rather than evidence that responsible AI or analytics systems automatically improve university performance. Its practical effectiveness, scalability and generalizability cannot yet be assessed because no datasets were generated or analyzed and the framework has not been applied in an institutional setting. Universities also differ in governance, organizational culture, digital maturity, infrastructure and national policy, all of which could affect implementation. Proposed future work includes expert assessment of the model, bounded pilots such as student-retention or program-performance projects, multi-institution case studies and cross-cultural comparisons. Suggested process indicators include data quality, transparency of analytical outputs, usability of decision support, traceability of managerial decisions and identification and mitigation of ethical risks. Until that validation occurs, the framework’s strongest claim is structural: it offers a testable map for making ethical governance integral to data-driven university decisions rather than a separate compliance exercise.

Subject of Research: A conceptual framework for ethical, governance-integrated data-driven decision-making in higher education institutions

Article Title: A design science framework for ethical data-driven decision-making in higher education

Article References: Karakaya, M. (2026). A design science framework for ethical data-driven decision-making in higher education. Discover Artificial Intelligence, 6(1), Article 1113. https://doi.org/10.1007/s44163-026-02216-w

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02216-w

Keywords: higher education, data-driven decision-making, ethical governance, responsible artificial intelligence, learning analytics, institutional analytics, university management, decision support systems, algorithmic bias, design science research, design, science