I am a Ph.D. candidate in Economics at the University of Pittsburgh. My research lies at the intersection of information economics, behavioral economics, and game theory, with a focus on how limitations in knowledge and control over the information environment shape communication and learning.
I will join Tsinghua University as a postdoctoral researcher under the supervision of Dr. Zhen Zhou.
This paper studies a behavioral model of persuasion where the receiver may misinterpret information designed by the sender. Instead of assuming perfect observation, we allow the receiver to confuse one message for another with some probability. This misinterpretation limits the sender’s ability to influence beliefs and disrupts the standard concavification method of Kamenica and Gentzkow (2011). We also study a second departure in non-Bayesian updating: naïveté, where the receiver fails to account for misinterpretation. Misinterpretation weakly harms the sender without affecting the receiver, but naïveté can benefit the sender at the receiver’s expense. In a binary example, we illustrate how misinterpretation reduces the sender’s optimal payoffs, and when naïveté lowers the receiver’s demand for information. We then extend the analysis to confirmation bias, where misinterpretation depends endogenously on beliefs and strategy. Our analysis complements prior work on non-Bayesian inference (de Clippel and Zhang, 2022) and symmetric noise (Tsakas and Tsakas, 2021), by focusing on structured misunderstanding in persuasion.
When a sender interprets data to persuade a receiver's belief about a state, the receiver faces two sources of content simultaneously: the strategic content - what the sender's choice of interpretation reveals about her private information, and the semantic content - what her interpretation literally says about the state. This paper develops a framework that accommodates consideration of both sources and characterizes a class of inferential behaviors. We axiomatize the receiver's inferential rule and show it admits a representation as a convex combination of strategic and semantic content, parameterized by a weighting function $\lambda$. The framework unifies cheap talk $(\lambda = 1)$ and narrative persuasion $(\lambda = 0)$ as special cases of a common primitive, and yields tractable comparative statics on how the inferential rule shapes equilibrium. When $\lambda$ partially weights both sources, the sender's language choice generates novel tradeoffs: semantic content anchors beliefs, which constrains feasible communication, and even negligible semantic sensitivity makes the cheap talk benchmark a knife-edge rather than a robust prediction. A receiver applying purely strategic inference in a cheap talk setting may be susceptible to narrative persuasion once messages with semantic content are available.