Search on Rugged Landscapes: An Experimental Study

This paper presents findings from a laboratory experiment on human decision making in a complex combinatorial task. We draw on the canonical NK model to depict tasks with varying complexity and find strong evidence for a behavioral model of adaptive search. Success narrows down search to the neighborhood of the status quo, whereas failure promotes gradually more exploratory search. Task complexity does not have a direct effect on behavior but systematically affects the feedback conditions that guide success-induced exploitation and failure-induced exploration. The analysis also shows that human participants were prone to overexploration, since they broke off the search for local improvements too early. We derive stylized decision rules that generate the search behavior observed in the experiment and discuss the implications of our findings for individual decision making and organizational search.