Some Thoughts on the Use and Misuse of Statistical Inference.
A discussion of the abuse of techniques of statistical inference resulting from inadequate attention to the relationship between the model used and the real world problem being addressed, and inappropriate interpretation of conclusions in the light of that relationship. The basic principle underlying all statistical inference is that we attempt to distinguish between alternatives by comparing observed behavior with that predicted by predictive models of those alternatives. The use of predictive models which do not describe the behavior of the alternatives between which we wish to distinguish is a clear violation of that principle. The principle is illustrated in a discussion of sampling from an urn. The technique of causal inference through partial correlation analysis is discussed as an example of the violation of this principle. On the surface, this technique appears to have wide applicability in analyses in support of policy studies, but further examination shows its applicability to be somewhere between highly questionable and totally specious. 17 pp.