Higher-order interactions at scientific conferences influence team formation

TL;DR

This work provides a formalization for the notion of group interaction over time by defining a taxonomy of synchronous and asynchronous group interactions and finds evidence that all interaction types described in this taxonomy are highly significant for team formation.

Abstract

Cooperation enables teams to solve complex problems that one individual alone cannot address. In science, collaborative teams have become the predominant way through which progress is achieved. These scientific collaborations arise though various mechanisms, among which interactions at conferences. The Scialog conferences, which comprise a series of small, interdisciplinary scientific workshops held over several years, are an ideal laboratory to study the network mechanisms leading to team formation. Building on existing work studying team formation from a pairwise perspective, we present a higher-order network perspective generalizing this framework. We provide a formalization for the notion of group interaction over time by defining a taxonomy of synchronous and asynchronous group interactions. We apply this framework to the Scialog case study using a stepwise selection logistic model and find evidence that all interaction types described in our taxonomy are highly significant for team formation. This higher-order network perspective provides a new framework for the study of collective behavior and group formation.

Figures (3)

• Figure 1: For groups of three, this figure shows the possible types of interaction that can occur. Type I: the three participants have interacted in the same group at least once (synchronous), type II: the three individuals have interacted with each other at different points in the conference but never at the same time, type III: two out of three possible pairs involving the three individuals have interacted, but never at the same time, and type IV: one out of the three possible pairs involving the three individuals have interacted.
• Figure 2: An illustration of the network structure of each conference series. Each network is aggregated over all conferences in that conference series. Visualized with XGI landry2023xgi.
• Figure 3: Interaction and collaboration: Panel A shows mean total scaled interaction time [scaled minutes] for people who ended up collaborating (left bars) or not (right bars) in a group of three. Data is aggregated across all four conferences and is shown for the four types of interaction defined in the main text. Error bars show mean values of the bootstrapped data with 95% CI. P-values of the Mann–Whitney U test: synchronous interaction, $1.04\times10^{-21}$; asynchronous clique, $3.3\times10^{-2}$; asynchronous wedge, $6.7\times10^{-6}$; asynchronous link, $3.1\times10^{-5}$. Panel B and C are a visualization of the logistic model results. Panel B shows the change in probability of collaboration as scaled minutes of each type of interaction increases. Panel C shows the change in odds ratio as the scaled minutes of each type of interaction increases.