Abstract

Analogical reasoning plays a significant role in the evolution of scientific thought. Not only is analogy extensively used in the early stages of investigation to demonstrate the plausibility of hypotheses, but in some fields, such as archaeology and evolutionary biology, it is often the strongest possible form of theoretical confirmation. This widely used form of reasoning, however, has seldom been subjected to rigorous examination by philosophers of science. Not surprisingly, there is a notable absence of standards for distinguishing between 'good' and 'bad' analogies in scientific reasoning. This dissertation proposes such criteria. ;The starting point is Hesse's principle that for a legitimate analogical argument, causal relations must obtain between the known and projected similarities in one domain, and there must be "no compelling reasons" for denying that "causal relations of the same kind" may hold in the other domain. The dissertation goes beyond this principle by incorporating causal relations within a broader framework of logical, explanatory and statistical connections. It then articulates the nature of these connections on the basis of contemporary accounts of scientific explanation, mathematical proof and statistical correlation. In effect, it formulates a set of models of analogical reasoning corresponding to the most prominent types of connections that occur in such fields as mathematics, physics, biology and archaeology. The dissertation uses these models to define and test workable criteria for a 'good' scientific analogy. ;The dissertation also discusses computational approaches to analogy, such as case-based reasoning . While CBR programs successfully perform a species of analogical reasoning, they work only in specialized and well-defined situations where knowledge of relevant factors is already established. Consequently, their applicability to exploratory science appears to be limited. ;The dissertation contributes to confirmation theory by extending the Bayesian approach to evaluating scientific hypotheses. Within the Bayesian framework, a recognized problem has been to justify the assumption that a hypothesis has non-negligible prior probability. A number of writers have suggested that analogies might be relevant to this aspect of the confirmation process. The dissertation affirms this intuition, and develops a novel procedure for providing the required justification