A Bayesian Learning Model Fitted to a Variety of Empirical Learning Curves

WHERE DOES TECHNOLOGICAL progress come from and what determines its rate of advance? In answering these questions, it is useful to decompose technological progress into the invention of new techniques and products and the improvement of existing ones. Roughly speaking, the economist sees invention as the result of research and development, and improvement as the result of experience-learning by doing. Because productivity growth on any single process is likely to be bounded, invention is the origin of long-run productivity growth. But the "level" effects of improvement on productivity have, in some activities, been found to be huge-on the order of several hundreds of percentage points. Thus understanding how the process of improvement works will help us better account for growth. This paper concerns itself with a simple model of one of the forces involved in improvement, namely, the improvement in productive efficiency that occurs as a joint product with output, or learning by doing.