Understanding Discordant Evidence in Science and Policy

This dissertation focuses on the problem of discordant, or contradictory, evidence and its use in policy. Evidential discordance refers to the situation when pieces of evidence produced by scientists point to contrary conclusions. Scientific evidence is always uncertain, but depending on the context, evidential discordance can points to a significant degree of uncertainty in the available evidence. In the simplest case, this means that neither the researcher nor the policymaker are able to discriminate between hypotheses to infer what the facts of the matter truly are. Which evidence should scientists prioritize, and who should policymakers listen to in these cases?

Philosophers of science have worked on the notion of evidence from a variety of epistemic, historical and methodological perspectives. Looking at this literature is like gazing into a fractured mirror: the pieces are there, but the reflection they produce is disunified. This dissertation develops a theory of evidence that shows how the fractured pieces hang together. By critically engaging with the philosophical literature on scientific evidence, from classic sources on scientific inference to more recently formulated Bayesian accounts of evidence amalgamation and measurement coordination, the dissertation provides a unified account of the central epistemic and methodological questions raised by uncertainty in scientific evidence and as well as responses to them.

In sum, the dissertation articulates the specific epistemic and methodological questions related to discordant evidence, including the difficulties discordant evidence poses for public policy, and proposes solutions to them. By fostering a connection with scientific practice, the framework of evidence that the dissertation has developed is relevant for philosophers, scientists and policymakers alike. The dissertation consists of four articles, of which one has been published, two are in review, and the fourth is an unpublished manuscript.