Research Projects

Abstract: The prevalence and consequences of overconfidence have been a topic of profound interest to scholars and management practitioners alike. However, in comparison, underconfidence—another type of miscalibration—has received scant attention. Across four main and six supplementary studies (N = 7,038), we investigate how people socially respond to overconfidence and underconfidence, and develop a theoretical framework that explains when and why we penalize each type of miscalibration. We propose that both types of miscalibration are generally socially punished due to their perceived intentionality, pro-sociality, and their harmful effect on group coordination. Further, we argue that the strength of the penalty varies critically depending on the form of miscalibration. Finally, we demonstrate that underconfidence has the potential to be at least as detrimental as overconfidence to organizations in an incentive-compatible study centered around an important business decision. Overall, this research provides insight into when miscalibration is likely to receive or evade social punishment and consequently persist, underscoring the importance of studying both underconfidence and overconfidence in organizations.

Abstract: Intergroup conflict is widespread, destructive, and often harmful to all sides. It has long been recognized that leaders play a central role in instigating conflict, and that holding a position of leadership increases individuals’ propensity to engage in conflict. In this paper, we investigate another crucial mechanism: followers’ tendency to support conflict-prone individuals as group leaders. We propose that, to better understand intergroup conflict, it is necessary to identify leader characteristics that both increase the likelihood of engaging in conflict and enhance followers’ support for such leaders. We examine one such characteristic—overconfidence, defined as confidence that clearly exceeds what is warranted by the situation. We find that individuals who exhibit overconfidence are more likely to engage in conflictual decisions in an incentive-compatible multi-round version of the Hawk-Dove game, and that this effect is partially mediated by myopic self-focus. Further, we show that leaders who display unjustified confidence receive stronger follower support from followers who experience more fear and greed—two emotions that are central to intergroup dynamics. Taken together, our findings help explain why leaders exhibiting unjustified confidence are prone to acting conflictually and when such leaders are most likely to gain group support, thereby shedding light on the psychological and social processes behind intergroup conflict.

Abstract: Across four studies (N = 1,891), we evaluate the circumstances under which people challenge others who exhibit excessively high confidence. We propose that two important processes shape people’s challenging behavior. First, people’s propensity to challenge depends critically on their own uncertainty, which we refer to as the relative uncertainty heuristic. Specifically, people are more likely to challenge excessive confidence, the more uncertain they are themselves. Secondly, we highlight that the public nature of challenging can lead to gendered challenging behavior. When challenging is publicly announced within groups, women are initially less likely to challenge than men. However, when challenging is private, gender differences disappear, consistent with anticipated negative social responses inhibiting women from challenging. Moreover, we observe that challenging has the potential to improve group judgments in an incentivized financial forecasting task. Taken together, our findings help explain when and why excessive confidence is challenged, and thus give deeper insights as to how groups can counteract the harmful effects of overconfidence.

Abstract: How language affects behavior is a question of enduring interest across fields such as psychology, sociology, and economics. Yet scholars have typically been limited in the tools available to conduct such studies, only able to vary a small set of aspects of language at a time to avoid ambiguous results. Here, we present a method based on explainable AI that enables interpreting the links between many different aspects of language and behavior simultaneously. We test our method in a large proprietary dataset of 20,880 responses to 180 distinct decision tasks and provide two further replications and extensions of the method, showing the robustness of its insights in manipulating humans’ risk-taking behaviors out of sample and its applicability to practical decision tasks faced by consumers. Further, we demonstrate the potential of our method to inductively uncover theories directly from the data, both recovering established insights from the literature and pointing to new avenues for exploration. As well as these theoretical advancements, our method has widespread applications in industry, such as reducing the number of conditions that need to be trialed in AB-testing.