Australasian Mathematical Psychology Conference 2019

Causality in social network research

Garry Robins
Melbourne School of Psychological Sciences, University of Melbourne

I discuss how we can make causal claims in social network research. This is important for we need some understanding of causal effects for successful network intervention. The issue particularly came to a head about a decade ago around the controversial results of Christakis and Fowler (2007), who argued that obesity is diffused across friendship networks. Their methods met much criticism, with arguments that homophily and influence were inevitably confounded in network studies, particularly with latent homophily effects. In studies of real network-based social systems, where it is impossible to measure everything, how can we rule out latent effects which by definition are unobserved? Yet, rather than give up on isolating network influence effects, we can be guided by early thoughtful work on causal inference, including by Sewell Wright, Ronald Fisher and David Cox. The resolution of this conundrum for networks may not lie in seeking the One True Study that will provide the final conclusion, but a research program that combines Big Data with small (or Thick) data. Across various research literatures on obesity, such a research program is effectively coming into place, with implications for how we understand mechanisms for obesity diffusion. For successful causal inference in social networks, we may need research that crosses scale, with both a microscope on the small and a macroscope on the large.