Monday, November 5, 2007 - 1:00 PM
26-1
Fisher, Yates, Nelder, and Thompson: The Development of Statistical Design and Analysis Concepts at Rothamsted.
This talk describes how agricultural research at Rothamsted has provided a stimulus to statistical research - for the benefit of biologists everywhere.
The research began when R.A. Fisher was appointed in 1919 to analyse the accumulated results (from 1843 onwards) of Rothamsted’s field experiments. The resulting techniques of analysis of variance and experimental design revolutionized field and laboratory experimentation, providing ways to avoid bias, improve precision and guarantee valid conclusions. Other new methodology included randomization tests, discriminant analysis and statistical genetics.
Under F. Yates, from 1936, the Statistics Department continued research into experimental design and multivariate analysis, including J.C. Gower's Canonical Variate Analysis, as well as devising survey analysis techniques for various UK Official Surveys.
The computer revolution began early at Rothamsted, in 1954 when it became one of the first institutes to have its own computer, an Elliot 401. Statistical computing continued under J.A. Nelder from 1968, with the development of the statistical program GenStat (now developed at VSN). In parallel, problems from Rothamsted and colleagues elsewhere in the UK agricultural and biological research community required - and inspired - further methodology and algorithms, including generalized linear models (for analysis of plant disease or infestation data), identification methods (e.g. for yeast species) and nonlinear modeling (e.g. for growth curves).
More recently problems from plant breeding, and now microarrays, have provided the stimulus for the establishment and further development of the REML algorithm for linear mixed models, by R. Thompson and colleagues at Rothamsted and inAustralia
In this talk I will describe some of these techniques, the applied problems that provided the original stimulus for their development, and their continuing relevance to current-day research.
The research began when R.A. Fisher was appointed in 1919 to analyse the accumulated results (from 1843 onwards) of Rothamsted’s field experiments. The resulting techniques of analysis of variance and experimental design revolutionized field and laboratory experimentation, providing ways to avoid bias, improve precision and guarantee valid conclusions. Other new methodology included randomization tests, discriminant analysis and statistical genetics.
Under F. Yates, from 1936, the Statistics Department continued research into experimental design and multivariate analysis, including J.C. Gower's Canonical Variate Analysis, as well as devising survey analysis techniques for various UK Official Surveys.
The computer revolution began early at Rothamsted, in 1954 when it became one of the first institutes to have its own computer, an Elliot 401. Statistical computing continued under J.A. Nelder from 1968, with the development of the statistical program GenStat (now developed at VSN). In parallel, problems from Rothamsted and colleagues elsewhere in the UK agricultural and biological research community required - and inspired - further methodology and algorithms, including generalized linear models (for analysis of plant disease or infestation data), identification methods (e.g. for yeast species) and nonlinear modeling (e.g. for growth curves).
More recently problems from plant breeding, and now microarrays, have provided the stimulus for the establishment and further development of the REML algorithm for linear mixed models, by R. Thompson and colleagues at Rothamsted and in
In this talk I will describe some of these techniques, the applied problems that provided the original stimulus for their development, and their continuing relevance to current-day research.
Audio File