On one hand, science is a paradigmatic source of good evidence, with quantitative experimental science often described in classical statistical terms. On the other hand, many philosophers, statisticians, and scientists who study the foundations of statistics have rejected its techniques as unfounded. How to resolve this tension? I suggest that one can better understand the evidential import of classical statistics in terms of probabilistic versions of conditions of adherence, safety, and sensitivity. Although these conditions have been employed with debatable success in the service of definitions of knowledge, I suggest that they are fruitful in understanding evidence in classical statistics. I also sketch how they lead to improvements in statistical methodology, along some lines that Deborah Mayo has suggested.
A weekly seminar of visiting speakers, work-in-progress talks, and discussion of recent work in epistemology, for faculty and graduate students. Please email Nick Hughes to be added to the mailing list.
Oxford Epistemology Group Convenor: Nick Hughes