Cognition and the Theory of Learning by Doing

Abstract:     
The standard model of utility maximization assumes that individuals make complete contingent plans before making a decision. This paper shows that when the environment is sufficiently complex the optimal rule entails learning by doing rather than contingent planning. An implication of this result is that spill-overs from learning by doing are less than perfect, a hypothesis central to many models of learning by doing, such as Rosen (1972). It is shown that one can derive learning curves that can be fitted to data. The model is also consistent with the observations of behavioral economics, such as Kahneman and Tversky (1979), that in the short run individuals may make biased decisions, though long run behavior is consistent with utility maximization. Finally, the model is applied to Becker (1968)'s deterrence model to show that the optimal amount of monitoring is bounded away from zero.

Keywords: Decision Theory, Learning by Doing, Incentives

Working Paper Series