Research

Effective climate action depends not only on technological effects and policy innovations but also on the leadership driving their implementation. Despite extensive research on solutions to mitigate climate change and reduce carbon emissions, progress often stalls at the implementation stage, prompting a crucial question: Which leaders are best equipped to deliver meaningful climate outcomes? Here, I show that political leaders with local embeddedness demonstrate advantages in advancing climate action. Analyzing two decades of city-level data in China, I leverage the central government’s leadership assignment system to reveal that cities led by embedded bureaucrats exhibit larger progress in low-carbon transitions. My findings indicate that such leaders are 1.5 times more likely to prioritize climate issues in public discourse, implement twice as strong emissions-reduction policies, and lead to significant reductions in air pollution and CO₂ emissions. I distinguish between two forms of embeddedness: emotional and professional. Emotional embeddedness, driven by personal connections to the locality, emerges as the primary driver of enhanced climate action, shifting leaders' preferences towards sustainable policies. In contrast, professional embeddedness, characterized by local administrative experience, plays a more modest role. This paper calls attention to the potential of appointing embedded leaders to accelerate progress toward carbon neutrality and sustainable development.

The ideological relaxation and fiscal autonomy of the 1980s initiated over three decades of bold economic reform in China, driving sustained growth. However, an analysis of over 50,000 bond issuance records reveals a sharp increase in local debt, which has progressively constrained policy choices at the local level. Our cross-sectional and dynamic examination demonstrates that China's once-nimble developmentalism has largely transitioned into debt-driven crisis management across most provinces. In 27 provinces, high debt servicing costs for bond obligations relative to local revenue have left authorities heavily dependent on central government transfers and new bond issuance to sustain basic governance. This reliance has significantly eroded the fiscal autonomy enjoyed in previous decades. Furthermore, we provide evidence that these growing fiscal constraints have hindered the capacity of high-debt provinces to pursue policy innovation, even in critical areas such as carbon neutrality. Our findings highlight the profound shift in China's local governance from local innovation to high dependence on the center, with significant implications for future economic and policy trajectories.

The leader’s preferences shape policy outcomes, however, the lack of accurate tools to measure the leader’s priorities, especially among autocrats, leads to overlooking these preferences as part of elite decision-making processes. To solve that, this paper introduces Novelty Probing, a new method for measuring the policy priorities of political elites across topics, and their influence on the same topics. This framework quantifies the novelty and influence of a leader’s ideas by utilizing semantic similarity between sentence embeddings to assess the deviation of their speeches from official propaganda, constructing indices for a leader’s novelty, and influence across policy topics. The Novelty and Influence indices are combined to create the Dominance Index, a metric for a leader’s ability to implement their novelty into policymaking. The paper exemplifies the method in the field of Chinese elite politics, by applying the Novelty Probing framework to Xi Jinping, using a corpus of Xi’s speeches and 179,823 China’s State-Council-issued communications. Hence, this study measures Xi’s policy novelty, level of influence, and dominance over China. To represent the usage of the measures, the paper provides five empirical results to study patterns of Chinese elite politics. First, Novelty Probing is used for a mini case study of the effect of Xi on health policy, by highlighting key speeches, and policy documents affected by these speeches. Second, the method is employed to reveal temporal patterns in Chinese elite politics. Third, Novelty Probing constructs a quantitative case for comparison between Xi and Li Keqiang, the Premier of China, and indices are used to study Xi’s consolidation of power after the 19th Party Congress. Fourth, the author conducts audience analysis for Xi’s speeches. Fifth, the framework aggregated evidence that indicates Xi has only minor dominance over foreign policymaking in China.

Conventional accounts of authoritarian politics argue that elites prioritize political survival, not ideology. In this paper, we challenge that view by demonstrating how ideology shapes elite competition in China. We argue that autocratic leaders use ideology to signal policy preferences and rely on personal networks to identify officials aligned with their ideological vision. We build a new dataset of over 50,000 speeches and 40,000 policy documents from local officials in China and develop a novel method to measure ideological alignment with Xi Jinping. We find that elite conflict revolves around socialism and economic issues. Local officials with personal ties to Xi who publicly align with his socialist ideology are more likely to advance in their careers. They are also more likely to implement socialist policies, with negative consequences for economic growth. These findings suggest that, contrary to dominant theories, ideology plays a central role in structuring elite politics under authoritarianism.

Hamilton’s Paradox highlights the moral hazard faced by local governments due to the implicit expectation of central government bailouts. Where the central fiscal authorities can credibly deny bailouts to the localities, local debt can be constrained. This paper sets forth a framework where moral hazard intensifies at both the central and local levels when financial repression eliminates pressure and policing from external creditors. In such cases, central authorities, even if they could discipline local authorities, may repeatedly raise debt limits for local governments out of two considerations. First, the central government prefers to avoid the deleterious policy and political effects of sudden stops in credit provision to local governments. Second, the central government aims to transfer fiscal obligations from its balance sheet to local balance sheets in order to maintain its international reputation as a reliable steward of the economy. Empirically, we demonstrate the benefits of financial repression to the central government by showing that rising government debt levels do not impact bond spreads, unlike in most developing countries. We then show that when local debts mature, Chinese local governments, backed by central approval, issue additional debt rather than impose austerity, regardless of outstanding debt levels. Third, by matching a comprehensive geospatial dataset of major floods with China’s provincial boundaries, we show that the central government does not provide substantial fiscal transfers to provinces affected by major floods but permits localities to borrow further for disaster relief and reconstruction. This pattern reveals that strict capital control protects China from external financial scrutiny, enabling local authorities to intensify debt accumulation at the behest of the central government.

The increased use of text-as-data methods has greatly influenced empirical research in the social sciences, thanks to advancements in large language models and natural language processing. However, an often neglected aspect is the curation of text corpora, which is essential for ensuring the validity of research findings. This paper addresses the need for a systematic approach to corpus curation in social science research. The paper presents a conceptual framework and practical tools for corpus curation, emphasizing detecting and removing textual and contextual outliers, assessing document quality, and validating metadata consistency. These steps are crucial for maintaining the integrity and reliability of text-based research. The proposed framework includes three stages: defining the corpus scope, ensuring text quality, and ensuring metadata quality. Our approach offers algorithms to support principled decisions at each stage, enhancing transparency and methodological rigor. The utility of these preprocessing steps is validated through simulation studies on political corpora, demonstrating significant improvements in downstream classification accuracy. This work underscores the importance of corpus construction as a critical component of the empirical research process and provides a comprehensive pipeline to support reliable and reproducible text-based analysis in social science.

Building on an independent database of 180,000 UN recommendations and a novel computational method, we present the most comprehensive study of HR debates to date. We develop a unique empirical model that measures topical density of discourse. This innovative instrument measures the discursive activity of UN HR bodies through a machine-learning textual analysis of their outputs, offering a dynamic map of evolving trends in human rights, both over time (diachronically) and across different mechanisms (synchronically) within the UN HR ecosystem. Leveraging this comprehensive dataset and sophisticated computational methodologies, we identify which protected groups are central to different mechanisms’ attention and highlight the major human rights issues that have witnessed significant changes in attention. Our research presents significant findings on the density of UN HR discourse and its implications for two major debates in the field of human rights law – HR proliferation and the structural critique of UN HR bodies.

Type inference methods based on deep learning are becoming increasingly popular as they aim to compensate for the drawbacks of static and dynamic analysis approaches, such as high uncertainty. However, their practical application is still debatable due to several intrinsic issues such as code from different software domains will involve data types that are unknown to the type inference system. In order to overcome these problems and gain high-confidence predictions, we thus present TIPICAL, a method that combines deep similarity learning with novelty detection. We show that our method can better predict data types in high confidence by successfully filtering out unknown and inaccurate predicted data types and achieving higher F1 scores to the state-of-the-art type inference method Type4Py. Additionally, we investigate how different software domains and data type frequencies may affect the results of our method.