Notable Advances in Statistics: 1944-1963
When peace returned in 1946, the government's wartime statistical groups split up and the statisticians entered diverse statistical occupations all over the U.S. and the U.K. As a result, strong university statistical research groups were founded in both countries.3 Training and research in science and engineering remained high priorities for the federal government. Statisticians received unprecedented support from federal agencies. There also was a phenomenal increase in publication of statistics textbooks and reference books, written to present branches of statistics that were created or expanded during the war.
In 1948, mathematician Claude Shannon introduces the term “bit” (short for binary digit) as the basic unit of information in computing – in other words, a 0 or 1; Shannon attributes the invented word to his colleague John Tukey. Universities began to introduce courses in sequential analysis and operations research. There were so many areas within statistics that statisticians began to specialize. The 7th edition of R. A. Fisher’s 1935 book The Design of Experiments appeared in 1960. By that time, statisticians had extended the design of experiments to include fractional factorial designs, incomplete block designs, response surface designs, and many more exotic designs. Excellent textbooks were available and experimental design was included in every statistics curriculum.
Sampling was an area of statistics that received high priority during the war years. The theory and methods of nonparametric statistics was an established research and instructional area for academic statisticians. Students had access to textbooks and handbooks on the general theory underlying nonparametric methods. The general theory of linear statistical models was available in textbooks; it encompassed regression analysis, analysis of variance for fixed effects, random effects, and mixed models, and analysis of covariance.
Another major area was multivariate analysis including the multivariate linear models, classification, discriminant analysis, cluster analysis, principle components, factor analysis, canonical correlation & regression, and time series analysis. Interest in the problem of multiple comparisons began in the 1950s. The jackknife method of bias reduction and uncertainty assessment, a forerunner of the bootstrap method, was being advocated. Textbooks on probability, decision theory, and mathematical statistics were available for use in undergraduate and graduate courses.
A. B. Hill, who had studied the incidence and prevention of civilian casualties in London during the war, collaborated with the physician R. Doll to publish in 1950 the famous paper that established the statistical link between smoking and lung cancer. That work stimulated a flurry of advances in biostatistical methods, especially for cause-effect studies. For example, in 1959, Jerome Cornfield showed how to quantify the minimum size of the effect that any confounding variable would have to have, in order to nullify the association already observed. In 1952, D. J. Finney published Statistical Method in Biological Assay, a key reference for biostatisticians. The Kaplan-Meier estimate (1958) based on censored survival data also became an important method in biostatistics. By 1960, the Cochran-Mantel-Haenzel method was an accepted biostatistical tool.
For dealing with multiple comparisons, Scheffe published his S-method and Tukey produced his multiple range test (T-method) in 1953. Dunnett published his method for analyzing multiple comparisons with a control and Duncan published his paper on both the multiple range test and the multiple F test in 1955. Scheffe reviewed and compared contemporary multiple comparison techniques in 1959.6
There was an important carry-over of WWII statistical research into the civilian world, as can be illustrated with some examples. After the war, statisticians disseminated quality control principles to industries around the world. Accepting and applying those principles helped Japan rise to become an industrial and economic powerhouse. In 1950, Genichi Taguchi introduced his statistical methods for quality improvement. The first randomized clinical trial was conducted in the UK in 1948. In 1950, R. A. Fisher was the first to use the term "Bayesian."9
The American Statistical Association (ASA) elected Helen Walker as its first female President in 1944. The membership in the ASA increased exponentially, doubling about every 17 years from 1914 to 1988. In 1940, ASA had 2500 members when the U.S. population was 132 million. By 1964, the ASA had more than 8000 members when there were almost 190 million people in the U.S. At that time statistical techniques were being used in many fields of study in addition to business, economics, and government. New sections of the ASA had been created in biometrics, statistical education, business & economic statistics, social statistics, and physical & engineering statistics. Several new statistics-oriented professional societies had been founded, including the Institute of Mathematical Statistics, the American Society for Quality Control, the Econometric Society, the Biometric Society, and the Psychometric Society. Among the new American journals were The American Statistician, Technometrics, and Biometrics.
The availability of multiple journals and many excellent books must have made it easier for statisticians at remote locations to follow developments and trends in the profession. However, with so many developments occurring at such a fast rate, we believe it unlikely that MSC’s understaffed mathematics department could keep up.
Statistical theory and practice were altered by the advent of computers, beginning at the end of the Second World War. UNIVAC, the Universal Automatic Computer, was introduced in 1952. It was the first commercial computer in the history of the world. Soon, all calculations relating to the federal census were completed by UNIVAC, work that was called “data processing.” The term “data science” was coined in 1960.12
Last Revised: 2021-06-15