Get the latest news
Subscribe to receive updates from the School of Data Science.

Network science provides a powerful tool for capturing and characterizing the structure of complex systems by representing relationships between pairs of entities. Yet much collective activity takes place in groups, and reducing those interactions to pairs can discard information about who participates together and how groups are composed. What does this simplification leave out, and when does it matter?
In this talk, Alex Gates, assistant professor of data science, examines these questions through two complementary studies of group composition and scientific collaboration. First, Gates introduces Pairwise-Adjusted Higher-Order Composition (PAHOC), a framework for testing whether the attribute mixing captured by a pairwise network is sufficient to reconstruct the composition of complete groups. For each group size, a maximum-entropy model reproduces the observed network’s attribute mixing while imposing no additional higher-order constraints. Comparing its predictions with observed group compositions reveals structure that the pairwise description does not reproduce and distinguishes this insufficiency from the additional reconstruction error introduced by pooling interactions across group sizes.
The research moves beyond the observation that projection discards information to ask whether the information retained is sufficient to explain the group-level patterns researchers seek to understand.
Gates then turns to the organizational context of scientific collaboration through ongoing research on university research centers and institutes. By linking detailed affiliations to researchers and publications, this work examines how collaborative relationships are organized within and across these units. It explores differences between collaboration concentrated around central researchers and collaboration organized into distinct, interconnected teams, as well as how these structures connect researchers across disciplinary and departmental boundaries.
Together, these studies consider how the choice of representation shapes our understanding of collective organization: when pairwise patterns provide an adequate description and when understanding research groups requires retaining more of their composition and organizational context.
Subscribe to receive updates from the School of Data Science.