[Popper and the problem of induction in epidemiology]

In this article we are discussing a few of the contributions by the Austro-British philosopher Karl R. Popper, one of our most influential contemporary thinkers, whose epistemological and socio-political theories have also penetrated the sphere of epidemiology. We are focusing mainly on the so-called problem of induction. We sustain, in line with Popper, that the scientific method does not use inductive reasoning, but rather hypothetical-deductive reasoning. Although the movement from the data evaluating a hypothesis to a conclusion on the latter goes from the specific to the general, that is, in an inductive direction, the induction does not exist as a reasoning process or inference. That is, there is no method that enables us to infer or to verify hypotheses or theories (we cannot explore all of the possible situations to see whether the theory stands up), or even to render them very probable. Besides, scientists look for highly informative theories, not highly probable ones. What we actually do is to propose a hypothesis as a tentative solution to a problem, to confront the prediction deduced from the hypothesis with actual experience, and evaluate whether the hypothesis is rejected or not by the facts. As theories cannot be verified, we can only accept them if they withstand an attempt to reject them. Consequently, the test of a theory consists of criticism or a serious attempt at falsification, that is, the elimination of error within a theory, in order to reject it if it is false. The objective is, thus, the search for true theories. For this purpose, the scientific method uses a systematic set of methodological (not logical) rules, that is, decisions. These methodological rules or principles can be summed up in two: [symbol: see text]be inventive and critical!, that is, propose bold hypotheses and subject them to severe tests of experience. Logic plays its role mainly by allowing us to deduce from a hypothesis the predictions to be confronted with the facts or evidence. This is applicable both to statistical inference as well as to causal inference. We argue that the criteria of causality used in epidemiology are none other than rules of the method designed for the same purpose: they are concerned with eliminating or reducing errors (chance, bias...) on testing a causal hypothesis. Consequently, the so-called ausal inference, the step from evidence to causal theory, is not a logical inductive or probabilistic process but rather a decision based on the evaluation of a causal hypothesis thanks to methodological rules such as the criteria of causality. We believe that the interest of the debate between the Popperian and the inductivist epidemiologists is not merely a matter of words, as, if we are aware that we do not operate inductively, that we cannot establish firmly hypotheses, not even affirm them probabilistically, we will presumably adopt a humbler attitude and look more for the errors in our theories than for their facile examples of confirmation.