Why should the tools of inductive inference differ from one empirical discipline to another? In a deeper sense, the tools do not differ. Econometrics shares its logical foundations with psychometrics and biometrics and, for that matter, with meteorology and even with experimental physics. Two or more sets of jointly asserted propositions about observable facts (not about mathematics or logic or ethics) are compared to determine which set is, in some sense, in better correspondence with facts. In Harold Hotelling’s terminology each set of propositions, or hypotheses, contains a subset which is ‘considered’ (tested) and its complement which is ‘maintained’ (assumed) for the purposes of an individual piece of empirical inquiry. Econometric usage (perhaps since Mann and Wald, 1943) attaches the name ‘prior’ to the maintained propositions, thus extending that old philosophical term to propositions which, although not subjected to test in the particular research piece in question, may well have been derived from previous observations. The ‘prior’ propositions give ‘specification’ (R. A. Fisher’s term) of a part, not all, of the properties of the studied phenomena — for example, the form of some functions and possibly the sign, the range, or even the exact numerical value of some of their parameters. To take a crude example, proportionality between two observable variables may be asserted a priori, but the choice between rival candidates for the proportionality constant (e.g., between the various estimates of the economists’ ‘velocity of circulation of money’) will depend on facts.