Memories of the Impact of Fisher's Work in the 1920s

Abstract

This article is based on a talk given on 19 September 1972 at a Symposium in Cambridge organized by The R. A. Fisher Memorial Committee under the general title: Contributions to Inference in Scientific Reasoning. While a number of the contributions were concerned with Fisher's work and ideas, others dealt with new developments in theory and application, inspired by his own contributions, either being put in hand at the moment or during the years since his death in 1962. In the more historical section of our discussions it seemed appropriate that something should be said about the background against which Fisher's early work was set and indeed on the impact which this work had in the 1920s on some of his contemporaries. In preparing a short presentation of this kind I soon realized that I could only speak with any authority about the impact on myself, and that this meant that my talk would be more autobiographical than I should otherwise have wished. However, as being one of only two or three mathematical statisticians alive today who had experienced the impact 50 years ago, I was persuaded that my reactions should be put on record. The years 1915-30 marked a period of transition. There had first been the development of what I might call the early Biometric School founded by Galton, Weldon and my father (who for convenience I shall call K. P.). That it played a very significant part in the history of our subject cannot be doubted; to realize this one need only compare the change in theoretical expertise and in the recognition of the part which statistical theory could play in applied fields which occurred in the 30 years between 1890 and 1920. The inspiration behind this earlier work had largely arisen from an urge to study the problems of heredity, and to measure its intensity with the contributors still, as it were, living in the near aftermath of that liberation of thought which followed Charles Darwin's publication of The Origin of Species. Rightly or wrongly, the founders of the Biometric School believed that the way to advance knowledge in this field and perhaps to catch a glimpse of natural selection at work was by the collection and statistical analysis of large samples of human, animal and plant populations, breeding freely under natural conditions. The statistical theory developed was therefore what can be termed a large-sample theory and to interpret the results a detailed study of the logic of inference was not really so compelling as it became later on. We must, I think, remember that during the years 1894 to 1930 University College London was the only place in the United Kingdom where statistical theory and its applications were taught to any depth. In the early years of the century E. T. Whittaker was teaching a little statistics in Edinburgh and there was Bowley's teaching at the London School of Economics, but this had a strong economic flavour. The main influence of Yule, who had been appointed a lecturer in Cambridge in 1912, was, I believe, in Economics and Agriculture; on the more mathematical side it was Eddington and Stratton who gave courses on Combination of Observations and Error Theory, mainly using the classical approach of the astronomers. As a result the graduate students, whether from this country or abroad, who wished to learn more of the techniques developed by the Biometric School, came, year after year, to University College, London.