STEM identity, the degree to which students see themselves and are seen by others as “STEM people,” has become a cornerstone concept in education science. Decades of studies have linked it to STEM literacy, long-term academic persistence, and career aspirations in science, technology, engineering, and mathematics fields. Yet the construct has suffered from a persistent problem: everyone measures it differently. Some researchers operationalize identity through competence, performance, and recognition; others focus on commitment and exploration. The same word can mean different things in different studies, and different words can mean the same thing. “Competence,” for instance, refers in some studies to a student’s actual ability to understand scientific principles, while in others it refers only to the student’s belief about that ability. Constructs as superficially similar as “recognition” and “self-concept” have both been used to capture self-identification as a STEM person. The result, the researchers argue, is conceptual confusion that undermines measurement validity, blocks meaningful comparison across studies, and slows the accumulation of knowledge in the field.

The solution proposed by Zhimeng Jiang and Xiufeng Liu of the University of Macau, together with Jennifer N. Tripp of SUNY Geneseo, comes from an unexpected corner of computer science: ontology engineering. Ontologies are formal, structured representations of concepts within a domain and the relationships among them, encoded in logical languages such as the Web Ontology Language (OWL) so that computers can automatically search, check, and infer how concepts are defined and related. Far from being mere taxonomies, ontologies are designed for both human interpretability and machine readability, allowing software to verify whether class hierarchies are logically consistent, whether certain relationships imply additional classifications, and whether categories overlap or contradict one another. Ontologies have been used to manage curriculum models, describe learning domains, and represent teaching competencies, but never before, the authors note, in an affective domain such as student STEM identity.

S-STEMIO is what ontologists call a bottom-level, or domain, ontology. It is grounded in a mid-level framework called the STEM Identity Ontology (STEMIO), which in turn rests on the Basic Formal Ontology (BFO), a top-level ontology developed under international standards and used in more than 700 ontologies worldwide. BFO divides all entities into two fundamental categories: continuants, which persist through time, such as a person or an attribute of a person, and occurrents, which unfold over time, such as a learning process. Within this scaffolding, S-STEMIO places every measured component of student STEM identity into its proper ontological home, distinguishing stable attributes (qualities), latent potentials that activate under the right conditions (realizable entities), and temporal phenomena (processes).

The construction process was systematic and demanding. The researchers began with a PRISMA-guided systematic review of quantitative and mixed-methods studies measuring student STEM identity, searching four databases and identifying 153 empirical studies spanning many STEM disciplines and student populations. From these studies they extracted every component used to measure identity, from widely used constructs such as competence, performance, recognition, and interest to more specialized disciplinary identities in physics, chemistry, engineering, mathematics, information technology, and data science. That extraction produced a list of 92 distinct identity-related components, each of which then had to be classified into the hierarchy of STEMIO according to its conceptual scope, definitional similarity, and ontological role.

The classification decisions reveal the depth of the analysis. Science competence, defined as a student’s level of understanding scientific principles, was judged to be an enduring property of a person, a realizable entity and specifically a disposition, and more precisely a mental function of the cognitive orientation variety, because it manifests only when a student engages in scientific problem solving. Science recognition, defined as recognizing oneself and being recognized by others as a “science person,” was treated differently: it is a stable internalized evaluative judgment rather than a conditionally activated potential, and was therefore classified as a quality, specifically a cognitive representation falling under the class of appraisal. Science interest, the desire or curiosity to think about and understand science, landed as a mental disposition, a latent tendency rooted in mental activities such as thinking, valuing, and desiring. Every component was then assigned a preferred definition in the Aristotelian format used across ontology engineering, in which a term is defined as a species belonging to a genus and distinguished by a differentia, ensuring non-circularity and semantic precision.

Two researchers performed the classification, definition, and relation-building independently, reaching consensus through discussion where they disagreed. Initial disagreement ran at 12 percent, concentrated on constructs that could plausibly fit multiple classes, such as recognition-related terms that some studies operationalize as including both self-recognition and social recognition while others treat as purely internal. The team resolved such cases by placing externally influenced terms under the broader appraisal class and restricting strictly internal ones to self-appraisal. A panel of six internationally recognized scholars in STEM education then reviewed the ontology in two half-day sessions, prompting targeted revisions: recognition constructs initially placed under self-concept were reassigned to appraisal, and several components labeled simply “competence” were relabeled “competence belief” to reflect that instruments were capturing subjective perceptions rather than objective skills. Finally, the team ran SPARQL-based machine queries to test whether the ontology could answer real competency questions, such as whether science identity should be modeled as a subclass of STEM identity, and to check for orphan classes, metadata gaps, and inconsistencies.

The results are illuminating for what they reveal about the field’s collective assumptions. The 92 components cluster into ontologically distinct classes that do not overlap, which means researchers are not merely measuring different facets of one construct; they are often operating from fundamentally different assumptions about whether identity is something students possess, something they enact, or something they develop over time. Three overarching orientations emerged. The trait-based, self-perception orientation frames identity as a stable internal construct, captured through self-recognition and recognition by teachers, peers, and STEM communities. The situated, contextualized orientation treats aspects of identity as latent until triggered by social, cultural, or task-specific contexts, and includes many of the most frequently measured components: competence beliefs, interest, self-efficacy, and sense of belonging. The performative, behavioral orientation views identity as something enacted through action and experience over time, represented by commitment, exploration, reconsideration of commitment, and behavior.

The distribution across these classes is strikingly uneven. Most components fall within the continuant branch, particularly under cognitive representation and mental disposition, indicating that quantitative research has focused overwhelmingly on students’ internal states: how they perceive themselves, what they believe, and what they value. Belief-related components proved especially prevalent, spanning disciplinary competence beliefs, value-related beliefs such as utility value and the importance of science, and emotional attachment beliefs such as affirmation, relatedness, and alignment. By contrast, the occurrent branch, which captures identity as an unfolding, time-based process of doing, reassessing, and becoming, is comparatively thin. Constructs such as external support, feedback, and reconsideration of commitment highlight identity as dynamic and transformative, but they are rarely operationalized in survey instruments. The ontology thus doubles as a diagnosis: the behavioral and developmental dimensions of STEM identity, theoretically crucial to understanding how identities form, remain underexplored by the field’s dominant measurement traditions.

The practical implications extend well beyond taxonomy. Because the ontology is machine-readable and publicly available on GitHub, researchers can retrieve the formal definition of any construct, check where it sits in the hierarchy, and verify whether a given survey scale genuinely measures what it claims to measure. The relational structure also offers guidance for statistical modeling: if a cognitive function such as science competence is realized through a cognitive process such as exploration, that ontological relation may justify modeling exploration as a mediator between competence and identity, aligning structural equation models with theoretically grounded semantics. For program designers and policymakers, the ontology can expose mismatches between stated goals and actual assessments, as when an initiative aims to strengthen students’ sense of belonging but measures only academic confidence.

The authors emphasize that S-STEMIO is not fixed. Ontologies are designed to be extended as new theoretical perspectives and empirical findings emerge, and future iterations could incorporate insights from qualitative research to capture the fluid, socially constituted dimensions of identity that resist easy quantification. The ontology’s basic structure is expected to remain stable, but its lower levels remain open to community refinement, positioning the resource as a piece of open-science infrastructure for the field. For a construct that underpins how millions of students come to see themselves in relation to science, the arrival of a shared, computable language may prove to be one of the more consequential developments in STEM education research in years.