Effects of peers influence on voting and health-risk behaviour at the population level without using link data

The Social network, understood as the structure of influence between individuals, is of major interest in the study of public opinion formation. However, it is impossible to know the complete social network at the population level. Sociological studies typically analyse the relation of opinion with the population socio-demographic variables, such as age, gender, ethnicity, religion or income. However, correlations between individuals are often treated as confounds rather than primitives. In this project we reverse these priorities and attempt to privilege the inference of universal inter-individual dependencies (common across multiple datasets) over understanding factors specific to any one opinion.
Our mathematical framework represents population opinions’ outcomes (specifically Brexit remain/leave and London Mayoral elections conservative/liberal data) as binary spins embedded in a high dimensional space that combines spatial location and social variables such as age, gender, income, etc., which is called Blau space. We model the spins configuration as an Ising like model. This approach allows us to infer the social network structure by fitting the parameters of a connectivity Kernel that tells us about the scale of decorrelation of the peers influence in the different socio-spatial dimensions of the Blau space. We also found that the peers influence accounts on average for a 30% of election outcomes.