Journal of Economic Literature
Vol. XXXIV (September 1996), pp. 1324-1330
The Discovery of the Residual: A Historical Note
National Bureau of Economic Research and Harvard University
THE CONCEPT OF total productivity and the notion that labor is not the only factor of production, that other factors such as capital and land should also be taken into account in a wealth of the nation calculation or a measure of its productivity, were discussed repeatedly in the literature of the 1930s. Two major strands of research came together, ultimately, in what was to become total factor productivity measurement and growth “accounting”: The first developed out of the national income measurement tradition, based largely on the work of NBER and what was later to become the BE A. 1 The second was influenced by Paul Douglas’s work on production functions. His work had been largely cross-sectional, but as time series data became available it was an obvious generalization to add trend-like terms to the function and allow it to shift over time. 2 This re-search tradition had a more econometric background and did not accept, necessarily, the various compromises and assumptions embedded in the national income accounts. It found a fertile soil in agricultural economics, spurred by the presence and teaching of Tintner at Iowa State University (Tintner 1944) and the later work of Earl Heady (see Heady and John Dillon 1961, ch. 1, for an early review of this literature). The two traditions came together in the work of Solow (1957), in some of my own early papers (Griliches 1960 and 1963), and especially in Dale Jorgenson and Griliches (1967). But they have also kept drifting apart.
The first mention of what might be called an output-over-input index that I can find appears in Morris Copeland (1937). 3 Once one started thinking about “real” national income and worrying how to deflate it, it was a relatively short step to the idea that the two different sides of these accounts (product receipts and fac-
1 This tradition and the data bases it developed, together with the development of Keynsian economics, also contributed to the rise of growth theory in the works of Roy Harrod, Evsey Domar, and Robert Solow. But that is a different story.
2 Douglas had been criticized earlier for not allowing for some kind of trend factor in his estimation equation, especially by Horst Mendershausen (1938), a criticism that Jan Tinbergen endorsed in his Econometric textbook, published in Dutch in 1941. Given the increasing availability of time series data and the general advice of that time to use “de-trended” data, it was not long before trend variables began to appear in production function estimation. The first one to do so, as far as I can tell, was Victor Smith in 1940, in his Northwestern Ph. D. dissertation on the productivity of the steel industry, followed by Gerhard Tintner in a number of papers based on data for U. S. agriculture (Tintner 1944 and 1946).
3 More thorough research may unearth even earlier references.