Is optimal recommendation the best? A laboratory investigation under the newsvendor problem

We investigate the impacts of the decision support system's recommendations on decision makers' psychology and decision behaviors under uncertain contexts where optimal solutions exist. As a representative of such contexts, the newsvendor problem is studied by using the method of laboratory experiments. Through providing an elaborately designed decision support system in Experiment I, we validate that the optimal recommendations help to alleviate human newsvendors' Pull-to-Center bias, i.e., the actual orders fall in the range between mean demand and optimal order that maximizes the expected profit theoretically, and decrease the bias asymmetry under two profit conditions (high or low). We also reveal that optimal recommendations can't eliminate the bias, as decision makers exhibit two competing psychological factors simultaneously when using the decision support system: algorithm aversion and regret aversion. Algorithm aversion persistently impedes them from following the superior recommendations, while regret aversion sometimes pulls them to approach to the recommendations driven by the feeling of experienced regret. Further, we redesign the decision support system in Experiment II and find that, although the conservative system recommendations are valueless compared with the optimal one, the well-designed radical system recommendations may eliminate the Pull-to-Center bias under the high-profit condition, through the interaction of the dominant regret aversion, dominated algorithm aversion, and the anchoring effect.