Research
Publications
2. “Portraits of Power: Facial Appearances and the Tacit Domain of Political Selection in China.”
with Junyan Jiang.
American Political Science Review (Conditionally Accepted)
As the epitome of modern, rational organizations, bureaucracies are often believed to select candidates based on rules and reason. We argue that intuitive-and even instinctive-assessments of candidates' external appearances sometimes underpin seemingly rational and calculated decisions. Using a novel, AI-based algorithm that learns and reproduces human assessments of facial appearances at scale, we examine how perceived facial traits influence the careers of over 4,000 mid- and senior-level Chinese officials. We find that officials who look more competent, trustworthy, and less aggressive enjoy significantly better promotion prospects and lower purge risk than their peers. Warmth-related traits (e.g., trustworthiness and non-aggressiveness) are especially valued at higher-level promotions and for male candidates. Additional analyses, including conjoint experiments with real officials, demonstrate that appearances' influence over selection preferences is comparable to performance or political connections. These findings challenge the prevailing meritocratic and relation-based theories of bureaucratic selection and highlight the role of impressions in the workings of government institutions.
1. “China and the Liberal International Order: A Pragmatic and Dynamic Approach.”
with Ruonan Liu. (2023).
International Affairs, Vol. 99, No. 4, pp. 1383–1400.
How can we make sense of China’s perceptions of and relationship with the liberal international order (LIO)? The majority of notable works on this topic have been written by foreign scholars who emphasize China’s discontent with, or challenges it poses to, the LIO, while Chinese scholars have either focused solely on academic debates or attempted to interpret official foreign policies. This article aims to provide a balanced analysis of China’s view of order by examining theoretical thought and policy practices from a Chinese perspective, drawing insights from both Chinese academic writings and government statements. We argue that there exists a perceptual difference between China and the West in understanding the postwar international order, with China emphasizing the functional part of the order and the West regarding it as based on its preferred ideological values. This perceptual difference has caused international suspicion about China’s potential intentions to overthrow the LIO. In practice, China’s approach has remained pragmatic, interacting dynamically with different subcomponents of the LIO. Although China’s recent interactions with the LIO appear more ambitious, internal and external constraints, insufficient capacity, and lack of universally appealing values are preventing the birth of a Chinese version of international order.
Working Papers
- “Agentic Framework for Political Biography Extraction.”
with Yifei Zhu, Jiangnan Zhu, and Junyan Jiang.Abstract:Producing large-scale political datasets demands extracting structured facts from unstructured sources, traditionally relying on expensive human experts and resisting at-scale automation. This paper develops and evaluates large language model (LLM)-based solutions to this bottleneck, focusing on elite biographies, one consequential class of political facts. We propose a two-stage “Synthesis--Coding” framework: LLM agents first search, filter, and curate evidence from heterogeneous web sources, then map curated inputs into structured records. We validate the framework across Chinese, American, and OECD political elites, benchmarking performance against human baselines using multiple state-of-the-art LLMs. We find that LLM coders match or exceed human experts when given curated inputs, and that agentic synthesis substantially outperforms human collective curation (Wikipedia) in open-web environments. We further identify a systematic bias: directly coding from long, multilingual corpora degrades extraction quality, and demonstrate that the synthesis stage mitigates this bias by compressing evidence into signal-dense representations.
- “The Disciplined Face of Power: Affective Signaling and Hierarchical Rule in China.”Abstract:
Existing theories of authoritarian durability focus on formal institutions of control, such as personnel management and propaganda, largely overlooking the informal behavioral norms that sustain elite cohesion. This paper argues that authoritarian control is sustained, in part, through a pervasive regime of “bureaucratic affect”—a set of unwritten yet strictly enforced norms governing the emotional expressions of political elites. We posit that these affective norms solve a critical principal-agent problem: emotional discipline serves as a credible signal of an official’s effort, reliability, and political conformity within an information-scarce hierarchy. To test this theory, I apply deep learning to an original dataset of over 20,000 video appearances of Chinese officials. My findings reveal a clear affective hierarchy consistent with a principal-agent dynamic. First, higher-ranking officials (principals) consistently display greater emotional calmness, embodying the "face of the state." Second, lower-ranking officials (agents) strategically display context-appropriate emotions—such as greater stress in high-pressure roles—to signal their effort and diligence. Finally, preliminary analysis shows that mastering this affective signaling correlates with positive career trajectories, suggesting it is a rewarded, informal criterion for political survival. These findings reveal a critical micro-foundation of hierarchical political control, demonstrating how affective regulation enhances state capacity and elite cohesion.
- “Winners Take All: Performance Targets, Bureaucratic Momentum, and Stratified Economic Growth in China.”Abstract:
Target-based governance systems are a defining feature of state capacity in China and beyond, yet the causal dynamics they trigger beyond immediate incentives remain underexplored. Moving beyond the static "promotion tournament" view, I argue that target fulfillment acts as a critical threshold that generates a "dual momentum"—both material and cognitive—stratifying local development trajectories. Using a regression discontinuity design on data from Chinese cities (2002–2023), I provide causal evidence that barely meeting a GDP target significantly accelerates a locality's subsequent economic growth. I unpack this black box through two distinct mechanisms. First, consistent with a resource allocation logic, success serves as a credible signal that unlocks investment and emboldens officials to ratchet up future aspirations. Second, using automated text analysis, I identify a strategic attribution mechanism: "winners" adopt a technocratic orientation focusing on structural reforms, whereas "losers" engage in blame avoidance by attributing failures to exogenous shocks. Crucially, this momentum is context-dependent: the growth premium is amplified when local agents outperform peers but attenuated when superiors themselves face less performance pressure. These findings reveal how targets generate a cumulative advantage, explaining the strategic stratification of local state capacity in hierarchical bureaucracies.
- “Spontaneous Outburst or Strategic Signal? Legislative Emotions in South Korea.”
with Haejo Kang.Abstract:Does electoral accountability shape not only what legislators say, but how they emote? Drawing on variation generated by a mixed-member electoral system, we argue that electoral accountability pressures systematically induce strategic emotional displays in legislative settings. Legislators elected in single-member districts face stronger personal accountability than party-list legislators, incentivizing visible criticism and vigilance as signals of responsiveness. We test this argument using a newly constructed multimodal dataset that integrates automated video-based emotion recognition, legislative transcripts, and biographical metadata for individual legislators, drawn from the full universe of plenary sessions—over 784 sessions (over 1,750 total hours)—of the Korean National Assembly between 2010 and 2025. Emotional expression is measured using an AI-based deep learning model applied to video recordings of parliamentary debates. We find a systematic relationship between electoral institutions and affective performance: legislators elected from single-member districts exhibit higher levels of adversarial emotional expression than proportional-representation (party-list) legislators. Furthermore, among district-elected legislators, those representing more electorally competitive constituencies exhibit higher levels of emotionally charged opposition. Exploiting within-legislator transitions from party-list to district seats, we further show that the same individuals increase their use of emotionally charged opposition as they move to district seats and their accountability conditions change. This within-person shift strengthens a causal interpretation of the relationship between electoral incentives and emotional performance. The paper advances research on democratic accountability by demonstrating how electoral institutions systematically shape emotional performance as a visible feature of legislative behavior, drawing on large-scale multimodal computational evidences.