George Edward Pelham Box (1919–2013) was one of the most influential applied statisticians of the twentieth century. A pioneer in time series analysis, experimental design, quality improvement, and Bayesian inference, he combined deep mathematical insight with an extraordinary gift for practical problem-solving. His textbooks became standard references across multiple disciplines, and his philosophical observations about the nature of statistical modeling continue to shape how scientists think about their work.
Early Life and Wartime Work
Box was born in Gravesend, Kent, and initially studied chemistry. During World War II, he was assigned to work on chemical warfare experiments, where his need to analyze experimental data without formal statistical training led him to teach himself the subject. This practical immersion in real problems became the hallmark of his career. He later said he was “made a statistician by the army.” After the war, he earned a PhD in statistics from University College London under Egon Pearson.
Bayesian Contributions
Box was an early and influential advocate of Bayesian methods in applied statistics. His 1965 paper with George C. Tiao, “Multiparameter problems from a Bayesian point of view,” and their landmark 1973 textbook Bayesian Inference in Statistical Analysis were among the first works to demonstrate how Bayesian methods could be practically applied to real problems in regression, variance components, and time series. Box and Tiao showed that the Bayesian framework provided natural solutions to problems that were awkward or impossible within the frequentist paradigm.
Box's most famous observation—“All models are wrong, but some are useful”—first appeared in a 1976 paper and was elaborated in subsequent work. This pragmatic philosophy aligned naturally with the Bayesian approach, which treats models as tools for encoding assumptions and updating beliefs rather than as claims about ultimate truth. Box emphasized the iterative cycle of model building, criticism, and revision.
“All models are wrong, but some are useful.”— George E. P. Box, “Science and Statistics” (1976)
Time Series and Forecasting
Box's collaboration with Gwilym Jenkins produced Time Series Analysis: Forecasting and Control (1970), one of the most influential books in applied statistics. The Box-Jenkins methodology for ARIMA models became the standard approach to time series forecasting in economics, engineering, and the natural sciences. While the original formulation was largely frequentist, Box later developed Bayesian approaches to time series analysis that leveraged prior information about model parameters.
Experimental Design and Quality
Box made fundamental contributions to the design of experiments, including response surface methodology and evolutionary operation (EVOP). His work on robustness—how sensitive statistical procedures are to departures from assumptions—was also deeply influential. He was a major figure in the quality improvement movement, working closely with industry and arguing that good experimental design was the key to scientific and industrial progress.
Career and Legacy
Box spent most of his career at the University of Wisconsin-Madison, where he founded the Department of Statistics in 1960 and later the Center for Quality and Productivity Improvement. He was married to Joan Fisher Box, daughter of the great frequentist R. A. Fisher, which added a piquant personal dimension to the Bayesian-frequentist debate. He published prolifically until late in life and was awarded nearly every major honor in statistics.
Born on 18 October in Gravesend, Kent, England.
Worked on chemical warfare experiments; learned statistics through practical necessity.
Earned PhD from University College London under Egon Pearson.
Founded the Department of Statistics at the University of Wisconsin-Madison.
Published Time Series Analysis with Gwilym Jenkins.
Published Bayesian Inference in Statistical Analysis with George Tiao.
Died on 28 March in Madison, Wisconsin, aged ninety-three.