Must a Process Be in Statistical Control to Perform an Effective Experiment? Taken from the Fall 2004 Newsletter
Abstract: The paper explains why in some designed experiments none of the factors seem to be significant in predicting the response. Deming already stated that a process must be in control with special causes accounted for before conducting an experiment. Otherwise, the special causes might inflate the variability in the response. The paper illustrates his point by comparing the results from two teams of students that performed the classical catapult experiment. The first team took a systematic approach and implemented procedures to control external and environmental factors. The second team did not do this. The first team found five main effects to be clearly significant whereas the second team results had some confounding of interaction terms that were not clearly significant. The author suggests holding the process in control as much as practically possible in order to have the smallest experimental error; thus, confirming what Deming suggested.
Keywords: Fractional factorial - Mean square error - Special cause - confounding - Aliasing - Cpk - Replication - Half-normal plot of effects