Willful Ignorance: The Mismeasure of Uncertainty

CHAPTER 12 TRUST, BUT VERIFY

It is 3 P.M. on a depressingly cloudy Thursday afternoon in early March of 1969.¹ I am a graduate student in the Harvard Statistics Department, somewhat bored as I listen to a distinguished invited speaker. Jerome (Jerry) Cornfield is a world-renowned leader in the development of statistical methods for epidemiology and medical research.² Today, he is discussing recent clinical trials being conducted by the National Cancer Institute, where he is currently a leading biostatistician. It is a subject unrelated to my thesis topic, so my mind wanders.

At some point, though, I am struck by a troubling thought. It occurs to me that randomized clinical trials of new treatments are really very crude. Given the inevitable improvement of biological knowledge, especially with respect to genetics, won't we soon have the ability to predict individual outcomes of medical treatments very precisely? Can't we expect that our categorization of disease states will become increasingly sophisticated? We will then be able to define subsets of individuals who are quite homogeneous; the probability of a cure, for example, within each such subset will be close to one or zero.

What I find disturbing is the possibility that statistical methods, based on probability theory, might become obsolete. Here I am about to embark on a career that might become irrelevant to the highly advanced science of the twenty-first century that I imagine will exist. At the end of the talk, we ...