Purposes and Strategies in Regression Analysis

Abstract. Data analysis remains a subject of practise and informal teaching. In the 1980’s, there were several attempts to formulate ingredients of a theory of data analysis, in which John Tukey took an active part. A branch of this field of statistical science consists of formalizing strategies for specified tasks. Such strategies can be used – in addition to training – for writing smart statistical procedures and programs for guided analysis. Furthermore, differ-ent strategies might be compared as to their success, as soon as there are evaluation criteria for the results – another aspect of the theory of data analysis. This report focusses on sketching a strategy for multiple linear regression as a starting point. It is dedicated to John Tukey who gave his comments to a manuscript on it 17 years ago. This version is based on a longer manuscript that dates back 17 years. I made a second version in 1989 and in March 2001, some editing in order to get ready for a workshop in Buenos Aires. The workshop was dedicated to the memory of John W. Tukey – and this was the reason to pick up this topic again. John had see