Talk: Sam Fletcher (Oxford)

Title:

Consistent Estimators and the Argument from Inductive Risk

(Joint work with Corey Dethier)

Abstract:

Do (non-epistemic) human values enter into the evaluation of scientific evidence? What has come to be called “the argument from inductive risk” for a "yes" answer is situated within Neyman-Pearson (NP) statistical testing (Douglas 2000; Rudner 1953). NP testing relies on accept/reject criteria and must balance the probabilities of false positives and negatives; in choosing the criteria, it’s argued, the scientist must employ non-epistemic values to consider the potential consequences of the two kinds of error. But not all of statistics has this feature. It’s open to users of Fisherian significance testing to follow Jeffrey’s Bayesian rejoinder (1956) and hold that, in reporting only the best estimate or p-value, they are only evaluating what the evidence favors. “Accepting” or “rejecting” a hypothesis is to be left to the decision-maker.