Abstract
The ways in which neoclassical production theory can increase the understanding of the determinants of productivity growth are investigated and compared with the original formulation of Verdoorn's Law. The relative statistical properties of level and rate-of-change versions of 3 distinct productivity functions are tested using a 2-stage least squares estimation procedure and US manufacturing data for the period 1949-1981. The 3 models are the simple Verdoorn model, an extended Verdoorn model, and a scale and learning model. Results indicate that, whether one uses a log-level or first-difference specification of the Verdoorn model, the omission of factor substitution possibilities and short-run productivity dynamics from the simple Verdoorn model leads to serious misspecification bias in estimates of the effect of growth on productivity change. Moreover, the Verdoorn formulation suffers from a failure to specify the economic processes through which growth affects productivity.