In his attack on my investigations of the method of path analysis Mr. Allan* commits a number of errors of logic which invalidate his criticisms. However, before describing these errors I wish to comment briefly on some of his more general observations. He says (note 7) 'It is as well to remember what we are doing when we examine a path analysis: we are trying to find reverse-operationalizations, i.e. what haue we really measured Allan may indeed undertake his path analyses in a condition of ignorance of the variables which he employs but O. D. Duncan typically does not, and neither do I. Not a little effort is being devoted to studies of the nature and reliability of variables which I hope to employ in path (and other) analyses. For example, in the sociological field I am currently engaged in a test-retest study of the key variables employed in the Oxford Social Mobility enquiry. And a considerable amount of theoretical and empirical work has been done on the shared meanings of informants in occupation grading enquiries (see especially Appendix A of Goldthorpe and Hope, 1974)Path analysis can, of course, be employed as a variant of factor analysis in the conceptual analysis of a variable (Heise, 1969) and in the analysis of hypothesized unobserved variables (Hauser and Goldberger, 1971; see also its recommended employment in association with canonical analysis in Hope, 1971, pp. 235fr) but its usefulness in substantive sociological enquiry depends on the prior interpretability of the participating variables. It would be tedious to attempt to illuminate the obscurities of Allan's general thesis in any detail. Nevertheless, it should be mentioned that his key concept, 'simplicity', has been the subject of much complex discussion by philosophers of science (see, for example, chapter 7 of Popper's Logic of scientific discovery) and cannot be advanced as a criterion for selection among theories as if its meaning were self-evident and beyond dispute. It is by no means clear, for example, why, starting from a neo-Marxist position (note 7) Allan should endorse a functionalist interpretation of the relations between income and occupational prestige (p. 9), and should suppose that he can infer from this which variables are adventitious, and which are essential, to his simple model. When we turn to Allan's detailed criticisms of my paper, which are presented on pages 198-201 of his article, we find that neither the arguments nor the conclusions are at all clearly expressed a circumstance which can lead only to confusion in this difficult field. His first point is that two passages which he quotes from my paper (p. 199) are contradictory. The first passage states a proposition having the logical form if a then q , that is if the independent variables are uncorrelated ( a ) then path coefficients are unambiguously interpretable ( q ). The second passage contains a proposition having the form/? therefore q , that is because, in a particular case, the dependent variable is (almost) uniquely associated with a single principal component of the set of independent variables ( p ), therefore the path coefficients are interpretable ( q ). Allan presents a case (ia) which both he and I agree satisfies a and q and he infers that my two propositions are contradictory. The confusion here is almost total; if my two propositions are contradictory then no example is needed to establish this. Clearly, however, propositions with the logical form if a then q and p therefore q are not contradictory.1 What leads Allan to suppose that they are contradictory or otherwise defective? The clue is to be found in his example ia. He appears to believe that this