Productivity, R&D, and the Data Constraint
Forty years ago economists discovered the “residual.” The main message of this literature, that growth in conventional inputs explains little of the observed growth in output, was first articulated by Solomon Fabricant in 1954 and emphasized further by Moses Abramovitz (1956), John Kendrick (1956), and Robert Solow (1957).’ The pioneers of this subject were quite clear that this finding of large residuals was an embarrassment, at best “a measure of our ignorance” (Abramovitz, 1956 p. 11). But by attributing it to technical change and other sources of improved efficiency they turned it, perhaps inadvertently, from a gap in our understanding into an intellectual asset, a method for measuring “technical change.” Still, it was not a comfortable situation, and a subsequent literature developed trying to “explain” this residual, or more precisely, to attribute it to particular sources (Griliches 1960, 1963a, b, 1964; Edward Denison, 1962; Dale Jorgenson and Griliches, 1967). The consensus of that literature was that, while measurement errors may play a significant role in such numbers, they could not really explain them away. The major sources of productivity growth were seen as coming from improvements in the quality of labor and capital and from other, not otherwise measured, sources of efficiency and technical change, the latter being in turn the product of formal and informal R&D investments by individuals, firms, and governments, and the largely unmeasured contributions of science and other spillovers. The prescription of additional investments in education, in science, and in industrial R&D followed from this reading of history as did also the hope and
expectation that the recently observed rates of “technical change” would continue into the future.