Abstract

It is common within politics for people to be accused of bias – and not without good reason. Many political actors are willing to say things they do not believe in order to push a particular agenda, while others are inadvertently biassed due to biases in their reasoning or the information they consume. This means many of the political information sources we encounter will make claims in support of particular parties, policies, politicians and ideologies irrespective of whether they are actually true. Determining who is biassed, how they are biassed, and accounting for these biases, is crucial to ascertaining political reality. Despite this, bias has only recently begun to be studied as a source characteristic. I will present a Bayesian Network model of how people can infer and correct for source bias when attempting to learn political information. I will also discuss an intriguing prediction of this model, which is that people exposed to testimony from two sources who consistently disagree with each other should polarise. The model therefore provides a rational explanation of mass belief polarisation when people are exposed to the same information. I present preliminary evidence that this model accurately predicts belief updating, and contributes to belief polarisation in both the lab and real world.