The relation between theory and application in statistics. (With discussion) (Q1914738)

The present paper is in effect a series of comments on the relation between theory and application. The view is eclectic, almost guaranteeing that many will find something to dislike about the paper! If there is a single implicit theme to the paper it is that discussion of different modes of inference, fascinating though it is, is less important than issues of problem formulation and perhaps this is a view on which there may be wider agreement. The paper concentrates on applications in science and science-based technology where the object is in some sense to gain understanding and to present conclusions for relatively public discussion and incorporation into the knowledge base of the field. That is, applications where there is an immediate decision to be made, including issues of personal decision making or personal beliefs, are exluded. In analyzing and interpreting data the use of models, implicit or explicit, is unavoidable. The results of descriptive statistics or exploratory data analysis are meaningless numbers or graphs unless interpreted as representing aspects of subject-matter interest and this requires some schematic view of the system under study. While methods without an explicit probabilistic base can be important, for most of the paper the usual route of formulating explicit probability models will be followed.