Commentary: 'Statistical Methods in Genetics' by Sir Ronald A Fisher

Abstract

Listen

Translate article

The title of this paper 1 gives little hint of why it has been chosen for republication as an epidemiological classic. Nevertheless, it can be read with benefit by any biologist concerned for the study of quantitative phenomena. It may lack direct epidemiological reference, but is valuable for what it portrays of interconnections of thought in the mind of one of the most innovative of 20th century scientists. Rather than influences of genetics on statistical science, these thoughts illustrate well the contributions of 20th century statistical thinking to the understanding of genetics, especially the outlook fostered by aspects of principles of experimental design that have become important in all fields of scientific research, not least in clinical science. Like most of his generation, Fisher must have long been familiar with the spinning of a coin or dice as an objectively fair way of scoring points or determining the order of play in a sporting contest or an indoor game. Yet, it took a great mind to see that carefully planned randomization of order of operations and choice of subjects should and could be adopted as a pillar of valid inference in the conduct of experiments in the biological sciences, alongside replication and sound estimation based upon reliably measured data. Despite early controversy, such randomizing had long before 1951 become accepted as a part of standard good practice in agricultural field trials and in many other fields of biological research. As epidemiologists advanced from mere description of recorded events to the search for chains of causation, they had to recognize the demand for quantitative inference and prediction, attainable only by interpretation of well-designed and executed experiments. Principles of valid experimentation needed to be learnt and adapted from other fields. In research on human beings, the randomizing of assignment of treatments to subjects had seemed impracticable until AB Hill and his colleagues showed how this could be effected for planned clinical trials while continuing to conform to the overriding requirements of medical ethics, and so preventing the entry of biases that, in the absence of randomization, are inevitable consequences of subjective judgements. A fascinating feature of this lecture is Fisher’s recognition that the random assortment and chromosomal recombinations that are now known to be inherent in the processes of gamete formation may, in some circumstances, legitimize a claim that records from a study in genetics can be regarded as logically equivalent to data that conform to the principles of a properly randomized experiment.