Discussion of “Principles of Exploratory Data Analysis in Problem Solving: What Can We Learn From a Well-Known Case?”—Rejoinder
暂无分享,去创建一个
Institute for Business and Industrial Statistics (IBIS UvA), University of Amsterdam, Amsterdam, The Netherlands We would like to thank the editor for organizing this discussion of our paper, and we appreciate the points brought forward by the discussants. Professor Vining’s example of the chemical company illustrates the problem that we aimed to address with our paper. The professionals responsible for designing this company’s problem solving methodology were tool experts—they were, probably competent, experts on (mostly confirmatory) tools for data analysis, but they were, apparently, lacking in understanding of the process of inquiry. For us, being an expert on the tools of data analysis is not enough for being called a statistician, as the latter for us also involves understanding of the process of inquiry, and the roles of exploratory investigations and confirmatory investigations in it. We follow professor Vining here in acknowledging professor George Box as our profession’s inspiring tutor on this thought. Dr. Simpson makes a point in place, warning the reader that in the actual process of problem solving, EDA and confirmatory data analysis (CDA) cannot be clearly distinguished—activities, techniques and pursuits related to these two are often thoroughly intertwined. Dr. Simpson’s assertion that problem solving cannot be clearly distinguished into studies requiring purely EDA and studies requiring purely CDA, is illustrated by the case of John Snow. The reader may have assumed Snow’s Grand Experiment (mainly CDA in intention and set-up) to be after the Soho episode which identified the pump, and more generally, the water system, as instrumental in the epidemic; but in fact, these two investigations were concurrent, with John Snow frequently traveling between the south and north shore of the River Thames. The distinction between an EDA phase which identifies the hypotheses, and a CDA phase in which they are tested, is often a simplifying reconstruction afterwards. For us, the reason for contrasting the concepts of EDA and CDA, is that both represent distinctive functions in the problem solving process (sometimes characterized as discovery and justification). Each function Address correspondence to Jeroen de Mast, Institute for Business and Industrial Statistics (IBIS UvA), University of Amsterdam, Plantage Muidergracht 12, Amsterdam, 1018 TV, The Netherlands. E-mail: j.demast@uva.nl Quality Engineering, 21:382–383, 2009 Copyright # Taylor & Francis Group, LLC ISSN: 0898-2112 print=1532-4222 online DOI: 10.1080/08982110903291757