Organizational Learning: An Experimental Study

In this paper we experimentally investigate decentralized, organizational learning. The main objective is to understand how learning members of an organization cope with the confounding effects of the simultaneous learning of others. We use versions of a simple "success-or-failure" game due to Blume and Franco [2003] as our stylized model of organizational learning. In this game, each player has two action choices and is unable to communicate or observe the action choice of their partner. Only one choice of actions by a pair results in a "success" (high payoff); all other action combinations lead to failure (zero payoff). The success or failure game is repeated for two rounds, but ends early if a success is achieved. Decentralization is captured by explicit constraints on players' strategies, for instance, players are unable to condition their action choices on the past actions of the other member of the organization. This game theoretic formalization of organizational learning together with the assumption of fully rational agents are novel features of this study. They allow for crisp predictions with regard to the frequency with which organizational members who have not achieved a success should switch their action choices as a function of parameters governing individual preferences, such as the rate of time preference, which serves as one treatment variable. As a second treatment, we compare the two-player organizational learning model with a baseline, one-player individual decision-making problem involving a choice that is analogous to the organizational learning model. Preliminary experimental evidence suggests that in the two-player organizational model, players do account for the confounding effects of the learning of others by switching actions less frequently than in the one-period, individual decision making experiment in line with the predictions of the theory. We are currently exploring, experimentally, the predictions in the case where the size of the organization is increases from 2 to 3 and the number of success outcomes remains fixed at 1