zoom link, öffnet eine externe URL in einem neuen Fenster

Meeting-ID: 280 519 2121
Passwort: 584190

Abstract:

For over 150 years, there have been debates between Bayesians, Frequentists, and hybrid views, called the "Statistics Wars". These debates involve complex disagreements about standards of evidence, the interpretation of probability, and the purposes of statistics. Contributors have come from across different disciplines: from economics to physics, philosophy to mathematics. Statisticians often have the sensible view that we do not need to make an absolute choice between Bayesianism or Frequentism. Instead, different techniques are better for different types of problems. However, this "pluralist" view has its own unanswered questions. For example, what constitutes a method being "better"?

In this talk, I provide (1) the background and (2) a brief introduction to a philosophical answer to that question. I call it Calibrationism, developed by Henry E. Kyburg and Jon Williamson. Calibrationism was also approved by Ronald Fisher in his later years.

First, I explain the most common contemporary Frequentist and Bayesian views in the philosophy of statistics. I also describe how Bayesian epistemology (the most popular view in the philosophy of statistics) is actually significantly disconnected from Bayesian statistics.

Second, I provide a brief formal introduction to Calibrationism. I describe how Calibrationism uses imprecise probabilities for a hybrid philosophy of statistics. Like Bayesians, Calibrationists agree that hypotheses can be more or less probable, and that we can sometimes reasonably use Bayesian statistical tools like Bayes Factors, Bayesian parameter estimation, and Bayesian evidence amalgamation. However, Calibrationists also argue that the application of these tools must always be governed by our information about relative frequencies, so Frequentist reasoning is fundamental. It provides a formal and systematic account of what makes a method "better" in a context. I finish by noting an important way that Calibrationism needs to be updated to reflect contemporary applications of statistics.