Bayesian network models of political polarisation

Cases of belief polarisation, where individuals' views become more divergent after consuming the same information, are often attributed to motivated reasoning. However, work outside political psychology has demonstrated that Bayesian network models of cognition can generate belief polarisation among agents whose only motivation is accuracy. These demonstrations have never been empirically tested within the domain of political psychology before. The goal of this research project, therefore, is to identify Bayesian network models which can plausibly explain real-world cases of political belief polarisation, and test whether these offer a better explanation than motivated reasoning accounts.

One defining feature of Bayesian network explanations of polarisation is their assumption that individuals do not believe sources of information to be wholly truthful, whether intentionally or incidentally. A key factor here is whether people perceive sources of information to be biased. An attribution of bias might arise for numerous reasons - they might think the source is motivated to spread misinformation they know is untrue, or that they sincerely believe false information due to their gullibility, for instance. One further goal of the project, therefore, is to enhance understanding of why and when bias is attributed to information sources.

Attributions of source bias could be the product of motivated reasoning - people attribute bias to people who say things they don't want to believe - or partisanship - people attribute bias to outgroup members and honesty to ingroup members. Or, they could be explained by people making reasonable inferences from their beliefs about the biasedness of the source and any relevant groups to which they belong, the biasedness implied by the content of their message, and how common misinformation and biased sources are in their political information environment. This would suggest that the information environment to which people are exposed, and their perception of it, is important for understanding polarisation, and might help explain why polarisation levels differ across countries, times, and political systems, something explanations grounded in motivated reasoning and partisanship struggle to address.

Linking environmental-level factors to individual-level cognition through the lens of Bayesian network approaches, in pursuit of a fuller understanding of polarisation, is the ultimate aim of this research project. It is hoped that this will work will establish findings that can guide depolarisation initiatives in the future.

The project will use a mixture of computational modelling work and experimentation. This research primarily considers literature within social psychology and cognitive psychology, but is also be informed by epistemology and sociological work.