Linton Freeman Award Presentation: Sunbelt, Corfu, 2007

People in networks: Models for the interaction of individual qualities and social structure

Garry Robins

School of Behavioural Science

University of Melbourne

ABSTRACT

What drives and sustains a social system? How can we differentiate the importance of individual factors, external structural constraints and endogenous social processes in the way a system evolves? If we study a system in terms of individual-level qualities alone, we ignore the possibility of network-based processes. Exclusive concentration on network topology, on the other hand, obscures the role of individuals in shaping the system. Attention to only one factor ignores the possible importance of interactive effects between persons and structure.

In this talk, I show how psychological and social processes – incorporating psychological cognitions, individual attributes, prescribed roles, and social structural effects – can operate, often interactively, to bind an organization together. An illustrative analysis of the top management echelon of a multinational company reveals that structural processes are still important even after controlling for all observed individual effects. Conversely, relational structures also crucially depend on individual qualities, factors that may be subtle and not publicly visible. In the organization examined here, strong connections that cross established boundaries – the “glue” that provides informal organizational cohesion – emerge from the way in which individuals resolve competing psychological identifications.

To make these inferences requires a statistical model encompassing both individual and structural effects simultaneously. Because structure is so important to social systems, models for network structure are the first step. I review the new specifications for exponential random graph (p*) models that enable plausible statistical modeling of observed social networks, often with an impressive capacity to reproduce network characteristics. The inclusion of individual-level variables in these models then permits inferences about the joint importance of individual and structural processes.

With this methodology, social scientists need no longer be constrained to the analysis of single factors or of single structures, one at a time. The desire to simplify complexity by examining features in isolation is natural and helpful. At some point, however, we need to study postulated effects together, to disentangle the structural and the individual, and to understand their interaction. For such studies, good theorization as well as careful design and measurement will be necessary. We now have to hand the means to test our hypotheses in the cauldron of real network-based social systems, and to make principled empirical inferences about the complex interplay of people within networks.