HMI Talk: Johannes Lenhard, Prediction, Bayes, and Rationality
For many decades, Bayesian statistics was philosophically prominent, but remained an outsider in statistical practice. This changed in the 1990s when Bayesian methods quickly acquired high popularity also in the sciences. In this talk, I analyze the 1990s turn and argue that it hinges on an exploratory-iterative mode of computational modeling that is oriented at prediction. However, the new mode undermines the established Bayesian rationale by changing the role of prior probabilities. From an expression of beliefs held prior to new evidence they are turned into a parameter that is flexibly adapted over the course of modeling. This change likely induces a re-start in the philosophy of statistics.