Procedural Innovation Can Enable Learning the Optimal Strategy in Dynamic Decision Problems — Experimental Evidence
Posted: 25 Mar 2020
Date Written: February 28, 2020
To effectively explore for useful and valuable novelty, scientists, managers, and entrepreneurs alike routinely extract knowledge from the environment by deliberately eliciting and then evaluating informative outcome feedback. Our study aimed to better understand this process which, if the decision environment is complex, might itself require exploring for novel cognitive representations of the search space and feedback-evaluation methods. Our experiment zeroes in on how effective exploration is achieved in a computer-simulated business environment conducive to misperception of outcome feedback. The challenge stemmed from inertia in the variables to be controlled and delay in the effects of actions taken, jointly resulting in interdependencies between decisions over time. While in general participants tended not to elicit and evaluate outcome feedback appropriately, a case-by-case analysis revealed that some had successfully met the challenge by designing a learning procedure based on an appropriately simplified cognitive representation of the policy space. This procedure enabled them to systematically search in the subspace of steady-state policies and progressively learn the optimal trade-off between the variables to be controlled. Probing for the role of incentives, a treatment involved a bonus scheme remunerating participants who achieved high long-term performance scores with up to 400 euros. The bonus scheme increased the rate of trying new trade-offs but not the effectiveness of exploration. Accordingly, no effect on long-term performance was found, even when the task could be repeated. Our results point to the challenge of enhancing the means for crafting appropriate problem-specific cognitive representations and evaluation procedures.
Keywords: Innovation, Learning, Exploration, Cognitive Representation, Dynamic Decision Making
JEL Classification: D01, D21, D83, D99
Suggested Citation: Suggested Citation