Bayesianism Versus Baconianism in the Evaluation of Medical Diagnoses

Abstract

Medicine is an excellent example of a field in which it is often necessary to make inferences under conditions of uncertainty, either because further delays in therapy would be too risky or because further tests or examinations would be too costly. In the words of Sardeman, 1964, p. 142, ‘If you did all laboratory tests to everybody you would never get finished, and nobody would be able to pay for it.’ But some uncertain inferences are less reliable than others, and it is important to be able to draw the right distinctions here. So how is diagnostic uncertainty to be graded? Two models for the evaluation of medical diagnoses are often found in apparent conflict with one another today. The first is avowedly Bayesian. The second is demonstrably Baconian in its logic though not normally conceiving of itself as such. The purpose of the present paper is to clarify the nature of the difference between these two models and to show that, when restricted to their legitimate roles, they do not conflict with one another.