William Gemmell Cochran: The Influence of Rothamsted and Designed Experiments on His Research

Abstract

SUMMARY Employment at Rothamsted Experimental Station and early involvement in designed experiments were major influences on the research of William Gemmell Cochran. Many of Cochran's life-long statistical interests, such as the analysis of count data and the design and analysis of sample surveys, are readily traced back to his Rothamsted experiences with designed experiments. A number of his early contributions to the design and analysis of experiments are reviewed briefly. career in statistics that spanned half a century. His research, reported in over 100 papers and five books, had a major impact on the statistical design and analysis of both experiments and surveys; this impact is still evident today. As William James once said, 'Genius means little more than the faculty of perceiving in an unhabitual way'. Cochran possessed this faculty, but it would be simplistic to ascribe to it all his accomplishments; fortune, circumstance and dint of effort played their parts. Cochran's early publications give insight into the interplay of these forces and their effects on the evolution of this pre-eminent statistician. The central theme that emerges from these publications is that Cochran's early exposure to, and involvement with, designed experiments at Rothamsted Experimental Station was to have a most profound and lasting influence on his research. Cochran's statistical career began in 1931, when, upon completion of an M.A. in Mathematics at Glasgow University, he enrolled for graduate work at Cambridge. His first two papers (1934a, b) are the only two in his bibliography that list St John's College as his affiliation, and the only two published in the Proceedings of the Cambridge Philosophical Society; however, they are distinctive in other, more interesting, ways. The first contains the statement and proof of what today is colloquially referred to as Cochran's theorem, a wellknown result that has led to generalizations and extensions by numerous authors and has been applied to distributional problems resulting from partitioning of variability in an analysis of variance.