Another Look at the Statistician's Role in Experimental Planning and Design

In recent articles in The American Statistician, Hooke (1980) and Hunter (1981) discussed the role of the statistician in the experimental effort. They emphasized that the statistician's most important contribution to research programs often lies in helping to develop the logic of experimentation. The experiences resulting from collaborative efforts of scientists, engineers, and statisticians at the Battelle Columbus Laboratories are consistent with these views. We have found that often the most substantive contribution resulting from the interaction of researchers and statisticians has been to bring about a better understanding of the logical issues involved in the study. In addition, we have observed that many times these contributions are made during the initial planning stages of the experimental program that precede formal consideration of statistical design and data analysis issues. Much of the statistical literature focuses on the contribution that randomization, factorial designs, and other formal statistical concepts can make to the logical development of experimental programs. Few authors have discussed methods for obtaining important qualitative and quantitative information during the initial planning stages of research projects. For example, important information can be obtained through the construction of logic diagrams that outline the proposed tasks in the experimental program. The exercise of creating such diagrams can serve as a catalyst to generate supporting evidence for the need for the tasks, to uncover logical inconsistencies in the experimental approach, or to generate new ideas. In some cases deterministic models can be developed and used for sensitivity, uncertainty, or dimensional analyses. These types of analyses often reveal important parameters and variables, and they often highlight assumptions that are critical to producing valid inferences from the data which will be obtained from the experiment. In many cases these preliminary analyses can result in