This study investigates a vexing question in political science: do people bias their beliefs and processing of political information due to a desire to believe things consistent with their political group (“partisan motivated reasoning”) or do they simply do so because their prior views lead them to believe certain things over others, even though they ultimately strive to accurately learn about the political world (“Bayesian updating”). Scholars now mostly agree that such an empirical test will only be successful if directional desires, or “motivations,” are experimentally manipulated, which, I argue, has yet to be cleanly achieved since previous manipulations included political information, which triggers both motivational and Bayesian mechanisms. This study seeks to address this issue. In particular, I focus on two motivations which are typically viewed as central to why people might identify with particular groups: a desire to feel better about oneself (a “need for esteem”) and a desire to fit in and be accepted (a “need to belong”).

These motivations are tested through a two-wave, pre-registered online survey experiment conducted with over 2,000 respondents in the US using Bovitz, which recruits subjects to their online panel through probability sampling. Wave 1 captured all the necessary prior beliefs and provided the basis of the experimental manipulations provided in wave 2 (completed about a week later). Subjects in wave 2 were randomly assigned information about their answers in wave 1. This information is designed to inform respondents that they performed poorly (low self-esteem) or well (high self-esteem), or that they answered questions differently (low fit) or similarly (high fit) from the rest of respondents in wave 1. There is also a placebo control condition, to provide a baseline for the analysis. Since deception was thus used for most respondents, all respondents were fully debriefed at the end of the survey.

Respondents then answered a block of manipulation check questions, engaged in a 10 round learning game, and answered questions on partisan social identity in the US. The learning game, adapted from Hill (2017), was composed of 5 rounds on one political fact incongruent to Democrats (unemployment under Biden vs Trump) and 5 pounds on another political fact incongruent to Republicans (government deficit under Trump vs Obama), with respondents receiving probabilistic signals about the accuracy of their beliefs in each round. Responses to the game were financially incentivized. The effect of the experimental treatments on respondents’ prior beliefs (captured in round 0), their factual learning throughout the game, their confidence in their answers each round, as well as any downstream effects on political identity and partisan affect, will then be compared, in accordance with the pre-analysis plan. By arguably providing the first proper test of important components of partisan motivated reasoning, this study has implications for how we understand and treat political learning, political identity and even beliefs more generally.