Zhongwei Xu

I show that credence backed by weightier evidence is not only more resilient to challenging evidence learned in the future but also more resilient counterfactually. Even if the evidence had been different in the past, such credence would have changed by a relatively small margin. I further show that such counterfactual resilience amounts to being less susceptible to a particular kind of epistemic luck—evidential-doxastic luck. We can utilise the susceptibility to evidential-doxastic luck to rationalise a kind of ambiguity aversion that cannot be otherwise rationalised. It can be rationalised as expected utility maximisation not relative to a set of probability functions in the actual world but relative to a set of probability functions across the specified close possible worlds.

With the help of the formal structure of Bayesianism, I distinguish between two different ways in which one’s testimony can be undervalued. While the modelling in the existing literature only focuses on the undervaluation of the trustworthiness of the testimony, the undervaluation of the relevance of the testimony has not been given due attention. The undervaluation of relevance cannot be modelled with Jeffrey conditionalisation as it is traditionally practised. But it can be modelled with Adams conditionalisation or Jeffrey conditionalisation if applied to the right partition. Distinguishing between the two conceptions of testimonial undervaluation opens the possibility that a person can be subject to double testimonial undervaluation or both testimonial undervaluation and testimonial overvaluation because of the same social identity.