Breaking the Spiral of Silence - with Yuen Ho, Revise and Resubmit, Journal of Political Economy
Self-censorship is pervasive in today's social climate, distorting political discourse and decision-making processes that either explicitly or implicitly take public opinion into account. What drives self-censorship, and how can misperceptions about public opinion be corrected? The Spiral of Silence theory offers insights, positing that people who hold views perceived as socially inappropriate tend to self-censor, generating a distribution of expressed views that is skewed towards appropriate opinions. If the attention paid to silence is limited, this can exacerbate self-censorship and create an equilibrium where only socially appropriate views are expressed and considered dominant. Conceptually, we develop a two-stage signaling model in which self-censorship and limited attention to silence jointly interact to affect beliefs and public expression. We then experimentally test our model predictions in a dynamic, unified experimental framework where college students discuss controversial topics together. Our results show that students with socially inappropriate views self-censor to a significant degree. Given the limited attention paid to silence in the status quo, self-censorship amplifies over time. We experimentally increase the salience of silence and find that this leads to both more accurate beliefs about public opinion and increases the diversity of views expressed. Because inference and expression are self-reinforcing, different levels of attention to silence lead to diverging equilibria.
Correlation Neglect on Social Media: Effects on Civil Service Applications in China - with Yixi Jiang, Ziqi Lu, The Economic Journal, 2025.
Social media, on which signals are often correlated, has become a primary source of information. How do people form beliefs when reading correlated signals online? In a field experiment on Weibo, we exposed Chinese college students to redundant negative posts about civil service jobs. Consequently, they developed a 0.16 standard deviation more negative view of these jobs and were 11% less likely to register for civil service exams, compared to the control group. We find the same effect in a supplemental online survey experiment, and the effect is moderated by attention. Our study demonstrates that correlation neglect can affect beliefs and real-life outcomes, including career choice.
How Awareness of Media Bias Affects Reaction to News: Field and Experimental Evidence from Microblog Users - with Juanjuan Meng, Xi Weng, Revise and Resubmit, Management Science
This paper examines how awareness of media bias affects news consumers' reactions to news. Combining data from Chinese Microblog users and an online experiment that varies whether news sources are disclosed, we find that news consumers are more likely to repost news with a political inclination inconsistent with the media outlet's general ideology, with causal evidence from the channel of strategic inferences about media bias. Heterogeneity analysis reveals that individuals with Bayesian updating and level-k thinking abilities exhibit stronger effects. We also demonstrate that simply reminding people of media bias can be an effectively method to nudge sophisticated reactions.
Selection and Sorting when Supervisors have Discretion: Experimental Evidence from a Tanzanian Factory - with Yuen Ho
Almost all firms rely on supervisors to select and motivate workers, but what are the costs and benefits of doing so? In partnership with a large garment factory in Tanzania, we implement a series of field experiments to examine supervisor discretion in the selection of workers for promotion to leadership positions. In a first experiment with workers, we randomly vary whether the role of supervisors is emphasized in the selection process when workers make application decisions. During the discretionary selection process itself, we then randomize whether supervisors face financial incentives based on the quality of the workers they select. Our results show that delegation crowds in supervisors’ private information about the leadership quality of workers. However, delegation also generates costs for firms. Supervisor choices are misaligned with the firm's objectives, as incentives reduce the likelihood that supervisors refer a worker they are personally connected to. Correspondingly, discretion is disliked by workers and reduces the quantity, although not the quality, of workers who apply for promotion. Importantly, the effects of discretion depend on the quality of the decision maker. Higher quality supervisors make better discretionary decisions to begin with. Furthermore, supervisor selection does not discourage workers from applying, and crowds in higher quality applicants, when the supervisors choosing are themselves of higher quality.
Inefficient Advising and Blame Concern - with Yixi Jiang
People tend to withhold advice when they fear being blamed for unfavorable outcomes, which hinders the spread of valuable information. To investigate how blame concerns deter advice-giving and to test the effectiveness of communication in alleviating blame concerns, we conduct an online experiment using a two-by-two treatment design that randomly varies whether advisees can punish advisors and whether advisors can send messages. We find that potential punishment from advisees deters advice-giving by approximately 30%. Additionally, we observe gender differences in the effectiveness of messages: for male advisors, messages reduce the perceived likelihood and severity of punishment, alleviating blame concerns and increasing advising rates, while the effect is not significant for female advisors.