Co-evolutionary behaviour selection in adaptive social networks predicts clustered marginalization of minorities

Human behaviour is largely shaped by local social interactions and depends on the structure of connections between individuals in social networks. These two dimensions of behaviour selection are commonly studied in isolation by different disciplines and are often treated as independent processes. To the contrary, empirical findings on spread of behaviour in social networks suggest that local interactions between individuals and network evolution are interdependent. Empirical evidence, however, remains inconclusive as social network studies often suffer from limited sample sizes or are prohibitive on ethical grounds. Here we introduce a co-evolutionary adaptive network model of social behaviour selection that provides insights into generative mechanisms by resolving both these aspects through computer simulations. We considered four complementary models and evaluated them with regard to emulating empirical behaviour dynamics in social networks. For this purpose we modelled the prevalence of smoking and and the network structure in response to changing societal and normative support for smoking. Our simulations corroborate empirical findings: we found a reduced prevalence of smoking. Remaining smokers were suffering from reduced influence in the network and were preferentially connected to other smokers. The analysis of partial models suggests that the feedback loop between local social interactions and network structure is indispensable for capturing empirically observed patterns of behavioural change. Importantly, these patterns have also been reported for other examples than smoking, i.e., spread of obesity, happiness and social exclusion. Our results thus suggest a generative mechanism for the co-evolution of individual behaviour and social network structure. We conclude by discussing implications for sustainability transitions, medicine and neuroscience.