Abstract
SummaryCertain revolutionary changes in medicine—measurement, chemistry, genetics—have led to recasting both the criteriology and the conceptualization of the terms of discourse. But advances along this path rest no longer on naive observation but intimately and inextricably involve modeling, that is, a system of inference which derives no immediate warrant from the primordial data of the senses. This system is not totally new in quality, since all “fact” involves interpretation of data; nor is it entirely new in having heuristic value in unearthing, or calling attention to, unsuspected evidence. But it differs in two major respects. First, it requires patterns of surmise which can no longer be immediately corroborated at all stages by naive means but rest more and more on internal theoretical consistency and confirmatory predictions to be verified by other sophisticated and indirect means. Second, it is leading to fundamental upheavals in even the broad perception of human biology, undermining well-established groupings and exposing certain classical fields of inquiry as meaningless. The problem yet to be confronted in this field is how an epistemology of the model (in this sense) is to be constructed. The model is useless (as the whole scientific endeavor is useless) if it has not a more economical structure than the totality of the raw data, even if the success of this economy is in some degree factitious (“robustness”). On the other hand, the model is also useless unless it has heuristic value and helps in adjudicating among competing interpretations (“power”). A good model will help also to distinguish relevant aspects of the data from the irrelevancies ascribed to error. The paradigmatic function of the model, then, defines certain “sufficient statistics” which condense the relevant features of the data. In fact a close analogy can be developed between the methodology of statistical inference and nosology. Both are necessarily tentative, and healthy development of both, in practice, is iterative; that is, the evidence and the model are mutually correcting and lead to advancement one of another