Article: Rigorous Uncertainty: Why RA Fisher is Important

Harry M Marks, Rigorous uncertainty: why RA Fisher is important, International Journal of Epidemiology, Volume 32, Issue 6, December 2003, Pages 932–937

The author discusses the philosophy of the Founding Father of modern statistics: the brilliant but controversial RA Fisher. Some of the concerns he had about political and economic power stifling principled scientific inquiry that motivated his approach, remain especially relevant today.

Fisher’s highly technical arguments about the nature of probability and likelihood are rooted in his more general concerns about the evolutionary and political importance of intellectual autonomy … strongly reinforced by his ideological critique of Soviet science during the Cold War.

Especially interesting is that Fisher’s ideas on randomization and what has been absorbed into “Evidence Based Medicine” are in conflict:

I have three goals: first, to lay out Fisher’s views on randomization, and explain where they fit within his broader notions of statistical inference; second, to place Fisher’s ideas about science and statistics in the English intellectual landscape between the Great War and the Cold War; and third, to ask why it was that Fisher’s ideas about randomization and uncertainty had so little influence on medical understanding, then or now.

Fisher’s broad view on scientific inference was to liberate “thinking people” from excessive reliance on authority:

Fisher’s insistence that any ‘thinking man’ could understand the ‘principles of statistical inference’ was not simply rhetorical. Running through The Design of Experiments is Fisher’s conviction that ‘the right use of human reasoning powers’ would free ‘intelligent people’ from a dependence on authority.

Regarding randomization in medicine:

While Fisher believed in the superior rigour of randomized designs, he did not insist on the hard line that emerged in medicine between randomized and non-randomized studies.

There are a number of other important issues discussed: Fisher’s critique of decision theory, his skepticism of both large centralized governments and corporate interests dictating what scientists can do, and the importance of maintaining a climate of intellectual autonomy:

For someone making heavy use of Cold War rhetoric to indict decision—theoretic approaches, Fisher was surprisingly even-handed in his analysis. In the US, he argued, the danger to independent thought did not come from the state but from the values inculcated by the modern corporation:

The author quoting from Fisher:

In the US also the great importance of organized technology has I think made it easy to confuse the process appropriate for drawing correct conclusions, with those aimed rather at, let us say, speeding production, or saving money. There is therefore something to be gained by at least being able to think of our scientific problems in a language distinct from that of technological efficiency.

I think Fisher would have eventually revised his conception of decision theory to be less critical, and accepted as a working model, the scientist as an agent with the goal of maximizing the information value of experiments (broadly defined). He would likely be at the forefront of the research into the intersection among Bayes, Frequentist, and Fiducial methods.

Related Reading

Leonard J. Savage (1976). “On Rereading R. A. Fisher.” Ann. Statist. 4 (3) 441 - 500 On Rereading R. A. Fisher

Taraldsen, G., & Lindquist, B. H. (2021). Fiducial Inference and Decision Theory. arXiv preprint arXiv:2112.07060.

Seidenfeld, T. (1992). R. A. Fisher on the Design of Experiments and Statistical Estimation. In: Sarkar, S. (eds) The Founders of Evolutionary Genetics. Boston Studies in the Philosophy of Science, vol 142. Springer, Dordrecht.

The thesis of my presentation is that Fisher linked experimental design and estimation through his technical account of (Fisher) Information. In particular improvements in experimental design … may be quantified by an increase in the information provided by estimates from experimental data.