material in writing and for comments on earlier drafts. 1 Regression with Just the Facts

Regression, the workhorse of econometrics, is usually taught with a mixture of factual calculations and a metaphysical appeal to a transcendent reality invoked to interpret the results of those calculations. The metaphysics and the calculations are so interwoven that many practitioners are under the impression that, without the metaphysics, the calculations have neither meaning nor validity. In some forty years of working with regression, I have found, however, that the calculations, or others quite similar, are much more helpfully interpreted without the metaphysics by simply sticking to the descriptive, factual statements they enable us to make. In doubting that formal hypothesis testing in the mold we have all been taught is the best or the only approach to the analysis of macroeconomic data I am surely not alone. Leamer (1983), Poirier (1988), Sims (1996) and others have expressed various doubts on this point. In this note, I want first to state the metaphysical assumptions in the form of a fable which will, perhaps, serve to keep the practitioner aware of their transcendental nature. ( In the fable, I have in mind primarily regression on economic time-series data, but it is not without relevance to cross-sections.) I will then show how the statistics dependent on the existence of the transcendent reality for any meaning-- statistics like standard errors of regression coefficients, t-values, and F-statistics-- can be replaced by others with essentially the same informational content but of a purely factua