00268 øa148167663d78e6679e602ee961c3b

270

Montgomery

experiments are often used in the design and/or development of new products, the improvement of existing products, new process design or development, or troubleshooting and improvement of existing processes.

The design of an experiment involves both art and science. The science involves the statistical and mathematical principles that lead to the development of the experimental design and the analysis methods for the resulting data. This aspect of the subject is well-known; for example, see Box, Hunter, and Hunter (1978), and Montgomery (1991). The art of experimental design involves both statistically-oriented thinking about the problem and the integration of all the non-statistical knowledge to deal effectively with issues such as defining goals and objectives, choosing appropriate experimental factors and response variables, deciding how each potential experimental factor will be treated, the logistical aspects of conducting the experiment, and so forth. These issues are not discussed extensively in the standard textbooks, and are the subject of this paper. Other usefiil reading on this generaÅ‚ subject include Hahn (1977) (1984), Natrella (1979), Bishop, Petersen, and Trayser (1982), Hoadley and Kettenring (1990), and Coleman and Montgomery (1993).

Montgomery (1991) presents a seven-step approach for planning and conducting experiments. These steps, summarized in Table 1, are the focus of this article. The first three of these steps constitute the preexperimental planning phase. There is an old saying that "80 percent of success in life is showing up"; and this statement certainly applies to designed experiments. If the preexperimental planning is done carefully, then the probability of eventual success is high; if it is done poorly, then the experimenter is at risk of failure.

Coleman and Montgomery (1993) discuss steps 1-3 in Table 1 extensively, including methodology for evaluating potential experimental factors, determining the role each factor will play, understanding the various sources of experimental error, determining how randomization will be accomplished, and so forth. That paper should be read as a companion piece to this one. We now tum to a discussion of the steps in Table 1.

Recognition of and Statement of the Problem

This may seem to be a rather obvious point, but in practice it is often not simple to realize that a problem requiring experimentation exists, nor is it necessarily simple to develop a elear and generally accepted statement of this problem. Furthermore, problem statements tend to be rather broad; perhaps, in many cases, too broad to serve as the objective of a specific experiment. This is particularly true if one is to encourage sequential experimentation. For example, the problem may be to develop a new process for producing XYZ.