Canonical definition
A collective is a bounded set of participants whose interactions are relevant to a shared domain, task, identity, or environment.
Claim status: This bounded-collective definition is an original CSI proposition informed by scholarship on collective constructs, levels of analysis, and dynamically constituted social systems (Morgeson & Hofmann, 1999; Kozlowski & Klein, 2000; Sawyer, 2005).
Defining characteristics
Membership is specified
The focal participants must be identifiable through a defensible inclusion rule.
Connections matter to the construct
Communication, coordination, interdependence, identity, or another interaction process must be theoretically relevant.
The collective exists within a setting
Tasks, roles, norms, technologies, institutions, and external events shape the meaning of interaction evidence.
The boundary applies over time
Membership and interaction may change, so the collective must be indexed to an observation window.
Why the boundary matters
The same participants may belong to several collectives at once: a project team, a department, a leadership group, and an informal community. Each boundary creates a different object of inference. An estimate of team cohesion cannot automatically be generalized to the department, and a department-level condition cannot be attributed to every subgroup. Multilevel theory requires researchers to specify the level at which a construct is defined and the cross-level relationships being claimed (Kozlowski & Klein, 2000).
Boundary specification also prevents CSI from treating all available data as though it belongs to one meaningful collective. Inclusion should follow theory and context, not merely data availability.
Membership and participation
Formal membership and observed participation are related but distinct. A person may formally belong to a collective while contributing little during an observation window. Conversely, an external stakeholder may participate substantially without being a formal member.
CSI therefore requires the researcher to state whether the boundary follows formal membership, active participation, task interdependence, network connection, shared identity, or another justified rule. Ambiguous or changing membership should be represented as uncertainty rather than hidden. This requirement is CSI-specific; it extends the broader multilevel need to align the focal unit, construct, and evidence.
Collectives and noncollectives
| Example | Collective status | Reason |
|---|---|---|
| Project team | Usually qualifies | Members share tasks, coordination demands, and continuing interaction within a bounded setting. |
| Online community | May qualify | Qualification depends on persistent membership or interaction, a shared environment, and a defensible boundary. |
| Executive leadership team | Usually qualifies | Participants have recurring interdependent responsibilities and collective decision processes. |
| People who viewed the same webpage | Usually does not qualify | Common exposure alone does not establish a theoretically relevant interaction boundary. |
| All employees in a corporation | Not automatically | The organization may contain many nested collectives; organization-wide inference requires a justified system-level construct and boundary. |
| A demographic category | Does not qualify by itself | Shared attributes do not establish a bounded interacting collective. |
These examples are applications of the CSI boundary criteria, not empirical findings about every instance of the listed social form.
Nested and overlapping collectives
Collectives can be nested, overlapping, and dynamically reconfigured. A team may contain subgroups, belong to a larger program, and share members with other teams. These structures can influence an inferred state through cross-boundary dependencies, subgroup divergence, and external pressures. Multilevel frameworks explicitly distinguish nested levels and contextual influences rather than treating them as interchangeable (Kozlowski & Klein, 2000).
CSI should preserve these distinctions rather than collapse them into one level. Where multiple boundaries are plausible, estimates may be produced at several levels and compared, provided each construct and validation strategy is appropriately specified.
Dynamic boundaries
Collective boundaries may change through onboarding, departure, role change, merger, reorganization, or shifts in actual interaction. A model trained on a stable membership assumption may become invalid when the collective changes substantially.
Responsible CSI systems should monitor boundary drift, record material membership changes, and avoid interpreting changes in state that may instead reflect changes in who constitutes the collective. The emphasis on time and changing composition follows research treating emergence as a dynamic process rather than a static aggregation (Kozlowski et al., 2013).
References and scholarly foundations
Morgeson and Hofmann (1999) — collective constructs arising through interaction.
Kozlowski and Klein (2000) — levels, emergence, context, and temporal processes.
Kozlowski et al. (2013) — research designs for dynamic emergence.
Sawyer (2005) — social systems dynamically constituted through interaction.
Complete bibliographic details are available in the Scholarly Sources registry.
Relationship to collective states and CSI
The collective is the focal social entity. The collective state is the latent property attributed to that entity at a specified time. Collective-State Inference is the process used to estimate the state from theory-aligned evidence.
Without a defensible collective boundary, the level of the construct, the meaning of the evidence, and the interpretation of the estimate remain ambiguous.
Research status
RA-003 is aligned with the current pre-submission theoretical manuscript. Future versions will add guidance for algorithmic boundary detection, uncertain membership, and empirical tests involving nested and changing collectives. Version 0.2 adds claim-level citations and explicit provenance without changing the canonical definition.
Preferred interim citation
Clark, E. D. (2026). Collective. CSI Reference, RA-003, Version 0.2. Collective-State Inference Research Program.
See the Reference version history.