Tukey, John (1915–2000)

John W. Tukey was a mathematician and statistician responsible for many innovations in data analysis. He was born in New Bedford, Massachusetts, and educated at home until he entered Brown University in 1933. After earning degrees in chemistry at Brown, Tukey entered Princeton University in 1937 to continue his study of chemistry. He began attending lectures in the Department of Mathematics, and, in 1939, received a Ph.D. in mathematics at Princeton. He remained there for the rest of his career as Professor of Mathematics, and later, he served as the founding chairman of the Department of Statistics. For most of his career, Tukey also held positions at AT&T's Bell Laboratories, where he worked on projects such as [Page 1048]the Nike missile system, the methods for estimating the depth of earthquakes, and the development of an index for the literature on statistics. He retired from both Princeton and Bell Laboratories in 1985.

Tukey served as a consultant for many clients, including the U.S. government. During World War II, he joined Princeton's Fire Control Research Office, where he worked on issues related to artillery fire control. Later, he applied his expertise in solving time-series problems to the issue of distinguishing nuclear explosions from earthquakes. As a consultant for Merck, he worked on statistical methods for clinical trials, drug safety, and health economics. His education-related consulting included work for the Educational Testing Service and on the development of the National Assessment of Educational Progress.

In 1950, Tukey was a member of an American Statistical Association committee that criticized, in a balanced report, the methodology used in Alfred C. Kinsey's research on sexuality. From 1960 until 1980, he worked for NBC on the development of methods for rapidly analyzing incoming election-night data. Later, he argued in favor of using statistical procedures to adjust U.S. Census enumerations.

Tukey was a proponent of exploratory data analysis (EDA), a data-driven approach that he thought provided a much-needed complement to inferential, model-driven, confirmatory data analysis methods. EDA emphasizes visual examination of data and the use of simple paper-and-pencil tools such as box-andwhisker and stem-and-leaf plots, both of which were developed by Tukey for quickly describing the shape, central tendency, and variability of a data set. These methods remain in use today and have been incorporated into most statistical software programs.

Tukey's work on the problem of controlling error rates when performing multiple comparisons using a single data set resulted in the development of his ‘honestly significant difference’ test. His creation of the ‘jackknife’ procedure for estimating uncertainty in a statistic whose distribution violates parametric assumptions is one widely recognized product of his work on the issue of statistical robustness. With this method, the variability of a statistic is estimated by successively excluding different subsets of data from computations. Because he viewed this procedure as a tool that is suitable for many tasks but ideal for none, Tukey coined the term jackknife. Other widely known terms first used by Tukey include bit (for binary digit), software, data analysis, and the acronym ‘ANOVA’ (to refer to analysis of variance).

...