A Collection of Articles by George Box
Edited by David M. Steinberg, Chair of the Technometrics Management Committee
Technometrics is proud to announce a special virtual issue commemorating the scientific career of George E. P. Box, who died in March 2013. George Box was a remarkably creative scientist whose celebrated professional career was always at the interface of science and statistics. Together with J. Stuart Hunter and Cuthbert Daniel, he was instrumental in launching Technometrics in 1959. Many of his articles were published in the journal. Therefore, we think it is especially fitting that Technometrics should host this online collection with some of his most memorable and influential articles.
George Box began his career in statistics as a soldier in the British army during World War II. Assigned to a unit working on antidotes to chemical weapons, he soon found himself designing, running, and analyzing hundreds of experiments. His experience was concrete evidence of the fact that statistics is integral to science and should be linked to scientific endeavor, which was a defining feature of Box’s prolific career.
After the war, he enrolled as a statistics student at University College London and worked with the Statistical Methods Panel at Imperial Chemical Industries (ICI). He earned his PhD from University College in 1952, working under the supervision of Professor H. O. Hartley. Gertrude Cox invited Box to visit North Carolina State University for one year in 1953. That remarkably fruitful visit led to a further invitation from John Tukey to head the Statistical Techniques Research Group at Princeton. Box spent three years at Princeton, leaving in 1959 to join the faculty of the University of Wisconsin-Madison and establish the department of statistics there. In 1985, together with Bill Hunter, he founded the University of Wisconsin Center for Quality and Productivity Improvement.
George Box made lasting and significant contributions to a number of areas, including design of experiments, time series, Bayesian statistics, process control, and quality improvement. Box-Jenkins methods became a buzzword for anyone working in time series, and the Box-Cox approach became the scientific standard for assessing the value of transformations. In addition to technical contributions, Box was deeply interested in scientific method and in what he saw as the vital role of statistics in science.
This virtual issue provides an opportunity to assemble some of George Box’s most important articles and to display many of his major research partners. The issue includes several sections, organized by subject matter. The clarity with which Box expressed his ideas and his keen sense for making his ideas available to practitioners give these papers lasting value. They are great reads, and all statisticians will learn from them.
Science and Statistics, Theory and Practice. This section includes Box’s R. A. Fisher Lecture, his address as president of the American Statistical Association, an article about the importance of practice in the development of statistics, and one of his last articles about statistics and innovation.
Modeling and Robustness. The issue has two articles on statistical modeling and robustness. The first considers data analysis from experiments with a hierarchical structure, in which time-yield profiles are observed at conditions laid out via a factorial experiment. The second, originally published as a Quality Quandaries column in Quality Engineering, discusses the impact of invalid assumptions in a statistical model.
Process Monitoring. The five articles in this section include three about feedback control. They lay out foundations that are crucial in reconciling Shewhart’s approach to statistical process monitoring with the ideas about automatic process control that developed in the engineering literature. One of the articles shows how these ideas fit into the general philosophy of Six Sigma. Another discusses CuSum and CuScore monitoring statistics.
Quality. There are two articles that focus on quality. Both consider aspects of experimentation for quality improvement and were stimulated by the quality engineering ideas introduced in the United States by Genichi Taguchi beginning around 1980.
Experimental Design. Box made pioneering contributions to experimental design. The four articles in this section include his defining paper with Stu Hunter on the 2k-p fractional factorial designs, his paper with Don Behnken introducing one of the most widely used response surface designs, his paper with Norman Draper on the joint consideration of variance and bias in defining a response surface design, and an article with Søren Bisgaard on the usefulness of the 12-run Plackett and Burman design.
Bayesian Statistics. George Box made a number of important contributions to Bayesian analysis. Unfortunately, most of those articles were not available for inclusion in this virtual issue. We include two more recent papers with novel applications of Bayesian ideas in the context of fractional factorial experiments.
Time Series. George Box and his colleagues, especially Gwilym Jenkins, made pioneering contributions to the use of ARIMA models and developed a strategy for fitting the models to data, based on iteration between tentative model, estimated model, and diagnostics for model criticism and improvement. He applied the ideas broadly, as will be seen in the four articles here, which relate to industrial process data, environmental monitoring, and economics.
The virtual special issue is available at Taylor & Francis Online.
Editor’s Note: The collection of articles here is restricted to those from journals now produced by Taylor & Francis; many other important papers were published in other journals.