Rothamsted Experimental Station 1919-1933
Rothamsted, 1914
In 1919 John Russell of the Rothamsted Experimental Station went to Cambridge in search a statistician who could make some sense of the mass of data which had accumulated during 80 years of crop trials. He was attracted to the unconventional approach of Fisher and at first hired him a temporary basis. A few days later Fisher was also offered a post at the Galton Laboratory, with the restriction that he could only teach and publish material approved by Karl Pearson. Fisher rejected the Galton post for the opportunity to do independent research in a subject he had already made his own: the statistics of small samples.
Fisher found himself among the company of agricultural specialists, including many graduates of Cambridge, and once again enjoyed the fraternal companionship he had missed since his college days. Professionally, he spent his happiest years at Rothamsted unimpeded by bureaucratic demands.
He also worked on genetical and evolutionary problems, including practical breeding experiments on poultry, snails and mice, from which he confirmed his theory of the evolution of dominance. He continued to publish papers on genetics and evolution, especially on variability, the evolution of dominance and mimicry. By the time Karl Pearson retired Fisher had established himself as the leading geneticist in Great Britain.
In 1929 Fisher was elected to the Fellowship of the Royal Society, joining the United Kingdom’s scientific elite and attaining scientific recognition of the highest sort. Fisher left Rothamsted in 1933 as head of a statistics department which attracted researchers and statisticians from all over the world to visit and learn about his new experimental design methods.
On display:
- Contract of appointment as head of the Statistical Department, Rothamsted Experimental Station, 7 November 1924. Fisher Papers. Series 24: Cuttings book re [early] career
- Newspaper clipping: ‘Sir Ronald Fisher: Aids to Agricultural Research’ / John Russell [Obituary] The Times, London [1962]. Fisher Papers. Series 26
- Photograph: Rothamsted, 1914
Scientific Advances
Studies in Crop Variation (1921)
Fisher quickly realised that new methods were needed which would enable sound conclusions to be drawn from experimenting with small samples of highly variable material limited by the number of units (plants, animals or plots.)
His ‘Studies in Crop Variation’ (1921) was the first of a stream of papers showing what could be done to improve experimental results. In the same year ‘On the "Probable Error" of a Coefficient of Correlation Deduced from a Small Sample’ developed the analysis of variance which provides a method for the estimation of errors of experimental results and for testing their significance.
The papers ‘On the Mathematical Foundations of Theoretical Statistics’ (1922) on maximum likelihood and ‘Theory of Statistical Estimation’ (1925) concerned small observational samples obtained from scientific experiments and clarified the distinction between sample statistics and population values.
On display:
- ‘Studies in Crop Variation’ / R.A. Fisher (1921). Fisher Papers. Series 2: Collected Papers, 15
The Latin Square and Randomised blocks
Latin Squares
Rothamsted had limited facilities and experiments were restricted to the testing of small samples. Fisher realised that many experiments had been too poorly done to prove anything at all and he set out to understand how such work could be done most efficiently.
Crop experiments had in past years been carried out using long rectangular plots of land. However, no two blocks could be guaranteed to be exposed to identical conditions. Fisher made it possible to circumvent the uncontrollable factors by introducing the experimental block lay-out based on the Latin Square.
For example, if the yields of four different varieties of wheat were to be tested (represented by A to D), the plots would be arranged so that each row and column contained one of the four varieties and was a block in its own right, decreasing the impact of variables such as soil fertility and exposure to weather.
He also proposed that several factors should be examined simultaneously, that different treatments be randomly assigned to different plots, and that statisticians should be consulted in the design of experiments rather than at the end of testing.
Many of Fisher’s most brilliant advances stemmed from the practical problems brought to him by experimental workers – leading to consideration of how replicated experiments could be analysed and how they could be better planned to achieve greater relevance and increased accuracy. Fisher never lost contact with real data and spent innumerable hours at his desk calculator testing and checking his ideas.
On display:
- Photograph: 5x5 Latin Square plots of an experiment near Beddgelert Forest to study the effects of exposure on different tree varieties, designed by Fisher and laid out in 1929. Courtesy: The Forestry Commission, 1945
Statistical Methods for Research Workers
In 1925 Fisher released his book Statistical Methods for Research Workers. At the time of its publication it did not receive a single positive review, critics being unimpressed with Fisher’s new philosophy, scope and method and the lack of mathematical proofs.
Yet the book would revolutionize agricultural research as it described methods for evaluating the results of small sample experiments and for the laying out of trials to minimise the disturbances due to the heterogeneity of soils and the unavoidable variance of biological material. Incorporating his previous work on sampling distributions based on the normal distribution, it also introduced Fisher’s ideas on randomisation in experiments as well as his reconstruction of regression theory.
The book revolutionised applied statistics, replacing the methods Karl Pearson had introduced at the turn of the century. No other book has had such a great impact on research methodology, not only in agriculture but also in biology and other natural sciences, and medical research such as drug response and dosage. It has been published in 14 editions between 1925 and 1970 and has been translated into six languages.
On display:
- Statistical Methods for Research Workers. 1st edition (Edinburgh: Oliver and Boyd, 1925)
The Genetical Theory of Natural Selection
The Genetical Theory of Natural Selection (1930)
Although he had been working on the publication for many years, The Genetical Theory of Natural Selection was not released until 1930. The book unified Darwin’s theory of natural selection with Mendel’s laws of inheritance, defined the new field of population genetics and revitalized the concept of sexual selection. It was dedicated to Leonard Darwin “In gratitude for the encouragement, given to the author, during the last fifteen years, by discussing many of the problems dealt with in this book.” Fisher’s wife wrote the entire book down at home, taking the place of the usual nightly family discussions for some time.
On display:
- The Genetical Theory of Natural Selection. 1st edition (Oxford: The Clarendon Press, 1930)
- Letter from Leonard Darwin to Fisher, 19 April [1930]. Fisher Papers. Series 24: Cuttings book re [early] career
- Letter from Julian Huxley to Fisher, 4 May 1930. Fisher Papers. Series 24: Cuttings book re [early] career
Sir Julian Sorell Huxley (1887-1975) was a fellow evolutionary biologist, eugenicist, and proponent of natural selection.
