Experience Early, Logic Later

Statistics educators have been trying to improve undergraduate statistics instruction for decades. Some progress has been made but the forces of the status quo are formidable. One of the most frustrating constraints relates to the economics of textbook publication: few publishers will accept a script that is much different from the current market. Another constraint is the human effort required by both instructors and students to blaze a new path. Moreover, the disincentives to teaching effort, especially at influential universities, are well known. In spite of these impediments to reform, a small group of reformers is motivated to keep trying. If students as a group had a keen interest in statistics, both teaching and learning would be more successful. In this paper, I want to encourage course designers and instructors to focus on the student motivation for the subject, even at the expense of shortchanging the student with the usual list of inferential tools. I will argue that guided immersion in real data-based problems, in contexts of interest to students, is a more effective way to produce useful learning of statistics basics than to present a logical sequence of techniques, even if the techniques are illustrated with applications as they are introduced. The organization of the paper is as follows: The first section considers an overview of the progress of statistics education over the last quarter century. Next, the style and content of textbooks is used as a proxy for the teaching style and course content of many undergraduate courses in statistics – that the advances in textbooks have not solved the pedagogical problems. The important role of context-based motivation, "experience-based instruction", is then discussed. Next, some suggestions are presented concerning the year-levels at which context-based instruction is appropriate, and the related issue of class size is considered. Three examples of context-based teaching of statistics theory are then outlined. The final sections of the paper discuss the implications of context-based instruction for both undergraduate and graduate statistics courses