Strategic management researchers use multi-armed bandit model simulations to study organizational learning in, for example, strategy implementation and investment choices. We criticize that previous studies used oversimplifications in the simulated task environment and assumed decision rules that neither reflect optimal nor actual behavior. By means of Monte Carlo simulations and conceptual considerations, we show that these issues severely limit generalizations of research findings. For example, decision rules intended to prescribe optimal behavior in previous studies are outperformed by a Bayesian bandit approach that we use in our simulations. Overall, our findings suggest that greater caution should be used when researchers derive practical implications from bandit model simulations. We use these findings to derive recommendations for future use of bandit models in strategy and organization science research.
Dieser Eintrag ist Teil der Universitätsbibliographie.
Search Authors in