Talk: Jan-Willem Romeijn (Groningen)

Title:

Lies, damned lies, and statistics

Abstract:

This talk showcases a number of insights from my book project. First I argue that statistical methods are inevitably relative. Much like Bayesian inference relies on the choice of a prior, frequentist statistical methods, here termed "Bernoullian", rely on the structure of sample space, exposing them to violations of the likelihood principle. I consider alternative approaches to induction, in machine learning and game-theoretic statistics, and argue that they do not escape relativity either. In the second part of the talk I consider what this means for the epistemic authority of statistical science. I discuss what the use of statistics commits us to, focusing on the role of models, and I offer a conception of relativized objectivity that combines insights from logical empiricism and pragmatism. Finally I argue that this conception of relativized objectivity can halt unbridled relativism and ward off a post-truth dystopia.