Learning by doing

Learning-by-doing is a concept in economic theory by which productivity is achieved through practice, self-perfection and minor innovations. An example is a factory that increases output by learning how to use equipment better without adding workers or investing significant amounts of capital. Learning refers to understanding through thinking ahead and solving backward, one of the main problem solving strategies. As Ying (1967) pointed out, this learning process is used in dynamic programming. The process is also used in strategic planning and chess. Doing refers to the capability of workers to improve their productivity by regularly repeating the same type of action. The concept of learning-by-doing has been used by Kenneth Arrow in his design of endogenous growth theory to explain effects of innovation and technical change. Robert Lucas, Jr. (1988) adopted the concept to explain increasing returns to embodied human capital. Yang and Borland (1991) have shown learning-by-doing plays a role in the evolution of countries to greater specialisation in production. In both these cases, learning-by-doing and increasing returns provide an engine for long run growth.

Recently, it has become a popular explaining concept in the evolutionary economics and resource-based view (RBV) of the .

The Toyota Production System is known for Kaizen, that is explicitly built upon learning-by-doing effects.

Learning by doing is often measured by progress ratios. This number represents the cost of production after cumulative production doubles. Dutton and Thomas (1984) survey various industries and find the ratio to be typically around 80%. Thus, if a good has progress ratio of 80% and costs $100 to produce after producing 100 units, when cumulative production reaches 200 units, it will cost $80 to produce. Formally, for some commodity, if Cost(t) is cost at time, t, d(t) is the number of doublings of cumulative output of the commodity in time, t, and a is the percent reduction in cost for each doubling of cumulative output (note: 1-a is the progress ratio), then we have Cost(t) = Cost(0)(1-a)^d.

...
Wikipedia