Philosophy of statistics
Synopsis
Probabilities and models are used to represent uncertainty, guide prediction, and support decisions, but these roles do not carry the same requirements. A coherent calculation must still be interpreted relative to the information, assumptions, and purpose of an analysis. Fictionalist Bayes treats the Bayesian agent as a constructed perspective, including information, probabilities, a model, actions, consequences, utility, and an updating and choice rule. Statistical pluralism studies how several imperfect representations can remain useful together. Work on probability asks which distinctions a representation must preserve to warrant a particular belief, prediction, or decision.
Specific topics
- Bayesian agents
- A Bayesian agent specifies information, probabilities, a model, actions, consequences, utility, and an updating and choice rule.
- Coherence and external warrant
- Internal coherence concerns relations within the specified agent, while external warrant concerns whether its conclusions should guide statistical practice.
- Statistical pluralism
- The interpretation of ensemble weights depends on whether they express model uncertainty, predictive performance, or the influence of predictive information.
- Probability and decision
- The adequacy of a probability representation depends on the distinctions required for belief, prediction, or decision.
Selected papers
Working papers
Bayesians Do Not Exist
Ensembling under Statistical Pluralism: A Taxonomy of Formulations, Diagnostics, and Reporting
Probability as a Normative Role