Toward a Metaphysics of Learning Analytics: Ontological Positioning of Data, Inference, and Normativity

Technical Summary: Toward a Metaphysics of Learning Analytics

Problem Statement
Despite fifteen years of rapid development since the inaugural LAK conference, the Learning Analytics (LA) field has largely neglected metaphysical inquiry. While epistemological and ethical debates have been vigorous, there is a scarcity of discussion regarding the ontological foundations of LA. Specifically, the community lacks a systematic derivation of LA's identity from its internal principles. Current definitions and practices often conflate the purpose of LA (enhancing learning) with its operations (data manipulation), leading to an ontological tension where normative judgments are implicitly embedded within descriptive processes. This paper addresses the fundamental question: "What is LA?" by examining the ontological status of the data it operates on, the agents involved, and the logical boundaries of inference.

Methodology
The paper employs descriptive metaphysics, following the framework established by Strawson [1959]. Rather than imposing a revisionary system or an ideal form upon LA, the authors analyze the existing conceptual structures and commitments inherent in LA's own definitions and practices.

• Source Material: The analysis is grounded in the official definition of LA provided by the Society for Learning Analytics Research (SoLAR), treating it as a privileged articulation of the field's self-understanding.
• Analytical Approach: The authors trace the logical implications of the definition ("collection, analysis, interpretation, and communication of data about learners") to derive ontological preconditions. They rigorously distinguish between descriptive propositions (is) and normative propositions (ought) to address the Humean is/ought problem within the context of LA.
• Scope: The inquiry focuses on the nature of "data," the identification of agents, and the logical limits of inference, explicitly excluding external normative justifications to maintain methodological neutrality regarding the ontology of the practice.

Key Contributions and Results

1. Ontological Refinement of "Data":
   The paper redefines the data manipulated by LA not as "given" facts, but as "Formalized Constituted Records."

   • Constituted: Data is not a raw, neutral given; it is constructed through conceptual and observational reduction (abstraction) from learner events.
   • Formalized: The records are structured to be operable by LA systems (collection, analysis, etc.).
   • Implication: There is no one-to-one correspondence between learning events and data; the meaning of data is determined entirely by the conceptual and technical frameworks that constitute it.

2. The Zeroth and First Principles:

   • The Zeroth Principle: The learner is the sole ontological prerequisite for LA. Learners are the subjects and producers of records; without the learner, no LA data exists. This is an intrinsic condition of data manipulation, not a normative goal.
   • The First Principle: The minimal legitimate ground of inference in LA is strictly limited to "formalized constituted records" about learners and their learning.
   • Logical Consequence: LA cannot logically derive normative "ought" statements (e.g., what constitutes desirable learning) from descriptive "is" statements (data records). Normative judgments must originate outside the LA system.

3. Identification of Eight Agents:
   The paper decomposes the LA process into eight distinct ontological agents to clarify responsibility and function:

   1. Learner: The subject and producer of data (ontological precondition).
   2. System Designer: Defines the framework and bears ultimate responsibility for the system's structure.
   3. Data Collector: Executes collection within the designer's framework.
   4. Data Analyzer: Performs analysis and selects methods.
   5. Data Interpreter: Derives descriptive insights and ensures theoretical relevance.
   6. Data Communicator: Transmits findings, selecting format and audience.
   7. Decision Maker: (e.g., learners, instructors, policymakers) who uses LA insights to make independent normative judgments and interventions.
   8. Knowledge Translator: Translates insights across different LA systems or contexts (e.g., researchers publishing findings).
   • Note on Teachers: Teachers are not treated as independent agents generating data but are subsumed under "Decision Makers" because LA's operational object is defined as data about learners, not data generated solely by teachers.

4. Critique of "Norm-Embedded LA":
   The paper identifies a critical ontological tension in practices termed "Norm-Embedded LA." These are systems where normative propositions (definitions of "desirable learning") are baked into the system design, data collection, or analysis phases prior to the generation of insights.

   • The Conflict: This conflation of purpose (enhancing learning) with operation (data processing) violates the First Principle. It creates an ontological inconsistency where the system attempts to derive "ought" from "is" within its own logic, potentially biasing analysis (e.g., focusing only on "at-risk" variables) and obscuring the distinction between descriptive findings and value judgments.

5. Positioning Adjacent Fields:
   The framework situates related fields ontologically:

   • Educational Data Mining (EDM): Positioned as the methodology for the "analysis" and "interpretation" phases within the broader LA ontological structure.
   • Learner Modeling: When models function as descriptive records, they fit within LA; when they act as normative structures that generate learning behavior, they deviate from LA's descriptive neutrality.
   • Learning Design: Exists in the normative realm outside LA, though its insights can be translated into system design guidelines via the Knowledge Translator.

Significance and Claims
The paper claims that this metaphysical system does not seek to restrict LA's practical utility or deny its educational goals. Instead, it aims to:

• Clarify Boundaries: Explicitly separate the production of descriptive insights (LA's domain) from the application of normative judgments (the domain of Decision Makers and educational philosophy).
• Safeguard Against the Is/Ought Problem: By establishing that norms cannot be derived from facts within the system, it prevents logical fallacies where data is mistaken for moral authority.
• Protect Agency: By positioning the learner as an ontological precondition and teachers as decision-makers outside the data loop, the framework protects the autonomy of both learners and educators from being reduced to mere data points or algorithmic outputs.
• Enable Critical Dialogue: The system serves as a stable reference point for the LA community to discuss epistemology and axiology without these debates becoming entangled with operational definitions.

The authors conclude that rendering these implicit assumptions explicit is not an act of imposition but a necessary step for the maturation of LA as an academic field, allowing for a clearer division of labor between the generation of evidence and the making of educational decisions.