Goal-driven information search biases create polarization and extremism during the accumulation of qualitative evidence

Information search biases contribute to the formation and exacerbation of polarization andextremism across many types of beliefs. These biases can be induced by different goals,manifesting even in rational decision strategies that maximize choice efficiency. In this paper,we explore how information sampling can be biased toward or away from extreme andpolarized information, focusing on the case where people consider qualitative multi-criteriainformation about their options. Specifically, we examine how the incentives of a choice taskcan lead to biased information search, while comparison or rating tasks can promote deeperinformation search and ameliorate or even reverse these biases. Across three studies, we (1)create ecologically valid information environments by having participants generate qualitativeinformation they might consider when making a judgment or decision, (2) quantify the strengthand extremeness of participant-generated pieces of information (attributes), and (3) track howparticipants sample this information in different goal conditions. We show that comparison andrating goals – where people rate the relative goodness of each option or rate them separately –can reduce the extremeness of participants’ beliefs and degree of polarization relative to choicegoals where they simply decide between the options. This benefit occurs even when participantssample from the same pool of information (i.e., have the same information environment) acrossdifferent tasks. Put together, our results suggest that comparison and rating goals can serve asbias-reducing interventions that promote more balanced information sampling across domainseven for qualitative and multi-criteria information.