Statistical models for social networks, such as exponential random graphs (ERGMs), have increasingly been used by organizational scholars to study the social interactions inside organizations, departments, and teams. While such models have been successful in providing insights about the local processes that underlie these interactions (such as homophily, reciprocity or transitivity), an additional interesting avenue for research focuses on the role of group-level contextual variables (such as the climate or composition of a team or organization) when considering a multitude of teams or organizations. In this paper we show how, in a team context, integrating team-level factors in a multilevel framework (i.e., a Bayesian hierarchical ERGM) enables us to answer questions about: 1) how team-level contextual factors might provide alternative explanations for the emergence of intra-team ties, and 2) whether any variation in local tendencies between teams might be dependent on such team-level variables (i.e., cross-level interactions). Using data collected among 103 MBA students who were grouped into 18 teams of 5–7 members to work on a project, we study the impact of members’ expertise and perceived psychological safety on advice seeking behavior. At a local level, we focus on the effect of these nodal attributes for homophily and differences in advice seeking and giving, while at the team level we incorporate the average expertise and the team’s psychological safety climate. Our results show that expert members are more likely to be the recipient of advice ties, but also that this effect is more pronounced in teams where the overall level of expertise is high. For psychological safety, we find that a high psychological safety climate impacts the advice relations among all members in the group, not solely those who themselves perceive a high level of psychological safety, suggesting that the team climate has an impact on all its members. This illustrates how a Bayesian hierarchical ERGM allows us to obtain relevant results even when studying small-sized groups.
Read full abstract