The failure of Frisch's vision: Frustrated attempts to integrate statistical and analytical approaches in econometrics, 1930-1960

When the Econometric Society was founded in 1931, its stated goal was "the advancement of economic theory in its relationship to mathematics and statistics." This was interpreted by Ragnar Frisch as involving a "unification" of statistical, mathematical, and theoretical methods in economics under the banner of Econometrics. Irving Fisher explained that such a unification was needed to implement the "scientific method" of analysis in economics, a desirable goal not achievable by strictly theoretical or statistical inquiry alone. Initial efforts in this direction were the business cycle researches of Frisch, Kalecki, and Tinbergen, which introduced dynamic analysis into economic theory.;Ways to implement Frisch's vision were explored throughout the 1930s and into the 1940s. The most influential of this work was the publication in 1944 of "The Probability Approach in Econometrics" by Trygve Haavelmo. Haavelmo there presented a unified approach to economic research based on the theory of probability and the principles of statistical inference.;However, Haavelmo's research program was only partially implemented. Work done under the auspices of the Cowles Commission emphasized the development of new statistical methods, but always accepted the underlying economic theory as a prior given. Econometrics proceeded to develop along the lines favored by the Cowles Commission, as can be seen in the early applied econometric research and the econometric textbooks.;My research examines the reasons why significant parts of Haavelmo's research program were discarded, and what effects this had on subsequent econometric research. This involves making a close examination of the discarded parts of the program (model construction and testing) and a comparison of applied quantitative research prior to and following the publication of Haavelmo's monograph. To understand why these parts of the program were discarded, I examine the historical development of mathematical statistics and statistical testing, as well as the development of probability theory. I found that there are problems of long standing in applying statistical testing methods, based on frequentist theories of probability, to scientific research. The Cowles researchers, when confronted with these problems, made the decision to set aside the problems of testing and model construction indefinitely to concentrate on solving problems of estimation and identification.;Previous histories have called the developments in econometrics during this time period a "probabilistic revolution," but that designation is based on the assumption that the program of Frisch and Haavelmo was actually implemented. My research shows that, to the extent such a revolution occurred, it was a revolution in technical methods only. The larger goal of a unified approach to economic research was lost with the discarding of the innovative aspects of the Haavelmo program.