TF-colloquium with Conor Mayo-Wilson

The department of Theoretical Philosophy of the Faculty of Philosophy in Groningen warmly invites you to a joint colloquium on
    "Statistical bargaining theory:  a framework for understanding inductive risk and values in science"
by Conor Mayo-Wilson from the University of Washington

"Statistical bargaining theory:  a framework for understanding inductive risk and values in science"

Many contemporary philosophers of science reject the so-called "value free ideal" for science, and instead, argue that scientists should consider the moral and political consequences of their actions when deciding which hypotheses to accept and reject.  I will argue that the most popular criticism of the value-free ideal -- the argument from "inductive risk" -- is not only unsound, but more importantly, misleading.  The argument misleads in two ways.    First, it conflates the way that moral and political values do (and should) influence the selection of general-purpose statistical tools with the way in which those general-purpose tools are adapted and applied in a particular setting.  The types of moral and political values involved in these two stages of science, I argue, are very different.  Second, the argument from inductive risk (and the recurrent discussion of the obligations of "the scientist qua scientist") encourages thinking of statistical hypothesis testing in a particular setting from a Bayesian decision-theoretic perspective, where the scientist performs an expected loss calculation to determine the appropriate threshold of evidence for rejecting the null hypothesis under investigation.   The Bayesian decision-theoretic perspective not only fails to describe actual statistical practice (because a scientist may use a classical test that fails to be a Bayes rule under any prior and loss function), but more importantly, it encourages scientists to attempt to collapse the often conflicting interests of different stakeholders into a single, numerical loss function.  I offer an alternative framework -- statistical bargaining theory -- that allows one to incorporate the diverse beliefs and values of different stakeholders into statistical decision-making, and I explain how the framework can be used both to evaluate general-purpose statistical tools and how those tools are applied in a particular context.