Title:
Rational Resistance to Correction: A Bayesian Analysis of the Continued Influence Effect
Abstract:
The continued influence effect (CIE) is a well-documented phenomenon whereby misinformation continues to affect people's beliefs even after they receive and accept a correction. Standard explanations attribute CIE to cognitive limitations or irrational tendencies. This paper demonstrates that CIE can occur even for ideally rational Bayesian agents, suggesting that the phenomenon is not merely a product of flawed reasoning but can arise from the structure of our informational environment.
We model both misinformation and corrective information as testimony from sources that are not fully trusted. To formalize partial trust, we develop an account of defeater propositions within Bayesian epistemology. We distinguish two types of defeaters: inaccuracy (the source is careless or unreliable) and indeterminacy (the source is manipulative or deceptive). While uncertainty about inaccuracy defeaters permits standard conditionalization, uncertainty about indeterminacy defeaters does not: the rigidity condition fails, requiring an alternative updating method that we introduce and develop.
Corrective information takes two forms: undercutting corrections target the source of misinformation (claiming it is inaccurate or insincere), while rebutting corrections directly contradict its content. We show that an ideal Bayesian can experience CIE after receiving either type of correction. This occurs even when the agent trusts the corrective source more than the source of misinformation: because neither source is trusted completely, both leave their mark on the agent's credences. We also show that timing matters for undercutting corrections about source manipulation: receiving such corrections before exposure to misinformation can prevent CIE in belief entirely, while corrections received afterward cannot.
This is joint work with Ed Pertwee (King’s College of London), and Mihaela Popa-Wyatt (University of Manchester)