Minseo Choi

We study whether individuals are willing to sacrifice informativeness to obtain advice that is more likely to confirm their prior beliefs. Subjects choose between two Blackwell-ordered advisors that also differ in their belief congruence. When there is a trade-off between informativeness and belief congruence, subjects choose the less informative but more belief-congruent advisor in 46.0\% of rounds. This choice rate drops to 25.3\% when the less informative advisor loses its belief-congruence advantage. To distinguish belief congruence preferences from accuracy concerns, we model advisor demand as a function of perceived accuracy and belief congruence. Our estimates reveal a significant average preference for belief congruence, with a marginal weight 39.6\% as large as that placed on perceived accuracy. In contrast, there is no evidence of directional distortion in belief updating upon receiving belief-congruent advice.

Abstract: We study a Bayesian persuasion game in which a decision maker (DM) incurs psychological costs from belief-disconfirming information and chooses whether to engage with the sender's experiment. We model preference-based confirmation bias as psychological discomfort from exposure to signals that conflict with the DM's prior beliefs. Whenever persuasion is feasible, the sender must commit to a strictly more informative experiment than in the Kamenica–Gentzkow (2011) benchmark. This increases the DM's "actual" welfare, defined as material welfare excluding the psychological cost term, when confirmation bias is moderate, while persuasion fails when the bias exceeds a threshold and the DM disengages. We then introduce processing-based confirmation bias, modeled as systematic misperception of signals. In isolation, it does not change equilibrium behavior or the DM's "actual" welfare, but when combined with preference-based bias it strengthens the disciplining effect for the DM. A dynamic extension shows that future engagement concerns further increase the sender's incentive to provide informative signals.

Abstract: This paper develops a theoretical model to examine how employment support programs for vulnerable groups affect a social enterprise's behavior and social welfare. We extend a standard constant elasticity of substitution (CES) production function by introducing capital utilization efficiency, allowing capital productivity to vary by worker type. This shows that non-financial support, such as training or technical assistance, can increase the relative employment of vulnerable workers. Assuming that social enterprises pursue cost minimization like for-profit firms, we introduce a firm-specific, exogenous baseline employment level and identify a threshold that governs additional hiring. Employment subsidies can shift this threshold, and we derive conditions under which they enhance social welfare. By incorporating capital utilization efficiency and baseline employment level, the model offers a refined explanation of policy mechanisms without assuming special motivations for social enterprises.

Abstract: We study auction-format preferences for a risk-averse seller in common-value environments with affiliated signals. A one-parameter power family of signal structures that nests the signal structure of the canonical uniform common-value model exhibits two important structural properties. First, the first-price equilibrium revenue equals the conditional expectation of the second-price revenue given the highest signal, so the latter is a strict mean-preserving spread of the former: every risk-averse seller strictly prefers the first-price auction. Second, the highest signal is a sufficient statistic for the common value, so English auction revenue coincides almost surely with second-price revenue. Combining the two, the linkage-principle inequality binds with equality on this family, identifying a knife-edge case in the affiliated common-value auction literature where the mean-only ranking is replaced by a strict second-order stochastic dominance ranking that reverses the seller's format preference. We discuss implications for auction design when sellers value revenue stability.

Abstract: We experimentally investigate how exposure to same-gender environments affects volunteering for low-promotability tasks (LPTs) in subsequent mixed-gender groups. We find that prior same-gender exposure induces an asymmetric response: men reduce volunteering by about ten percentage points in subsequent mixed-gender rounds, while women's volunteering is essentially unchanged. In addition, groups with more women coordinate more successfully. Averaged over all rounds, however, women do not volunteer more than men in mixed-gender groups, in contrast to canonical findings in the literature. In hypothetical assignment tasks, participants of both genders predominantly assign the LPT to a man, so the social-expectation channel emphasized in prior work cannot account for volunteering patterns in our context. We organize candidate interpretations into two belief-related considerations—the precision of beliefs about others' actions, captured by strategic uncertainty, and the level of empirical expectations about whether someone else will volunteer—and one preference-based consideration, gender in-group favoritism. These considerations fit different parts of the data but are not separately identified by our design.