Industrial Statistics: The Gap between Research and Practice 2007 Youden Address
Abstract: The article addresses the gap between research and practice in the field of Statistics. The article focuses on two sub-topics: Industrial Experiments and Statistical Process Control. In Industrial experiments, there are many papers on orthogonal two-level designs; however, we should focus more on nearly-orthogonal designs. Genichi Taguchi’s ideas were so widely adopted in the 80’s is because he proposed “one-shot” experiments that attempted to get rough answers in a timely fashion. However some of Taguchi’s ideas on experimentation and parameter design have limitations. For example, simplified designs where most problems were force-fit into an L18 design, ignoring interactions, development of ad-hoc methods of analysis based on S-N rations. We still use some of Shewhart’s techniques from the 1920s for Statistical Process Control when there are more advanced techniques like CUSUM and EWMA. We still use range charts to monitor variance. The S-chat (or some variation) should replace range charts. Using symmetrical control limits for S-, p- and c- charts when the lower limit is negative is a standard practice. Instead of using these limits and truncating the lower limit to 0, we should use non-symmetrical confidence intervals. Statisticians should incorporate engineering techniques (like root cause analysis) into their research. The main reasons why there is a gap between research and practice are: real problems tend to be messy and not readily translated into research topics that can be solved and published, lack of incentives for researchers, and a focus on short-term profits rather than long term growth.
Keywords: Youden Address - Industrial experiments - Statistical process control - Nearly-orthogonal designs - Control limits