Improving the gold standard in field experiments with multi-armed bandits

In field experiments, researchers commonly allocate subjects to different treatment conditions before the experiment starts. While this approach is intuitive, new information gathered during the experiment is not considered. Based on methodological approaches from other scientific fields such as computer science and medicine, we suggest a randomized adaptive allocation for field experiments in organizational research that is based on a Bayesian multi-armed bandit algorithm. By means of Monte Carlo simulations, we test the usefulness of this approach in a comparison with randomized controlled trials that have a fixed and balanced subject allocation. Our findings suggest that randomized adaptive allocation is more efficient in most settings. We develop recommendations for researchers and discuss limitations of our approach