Siyu Liang

How do authoritarian regimes make propaganda persuasive? This study evaluates the impact of foreign influencers in propaganda. Social media videos and state broadcasts from countries such as Russia and China often feature sympathetic Westerners, yet their effects on audiences remain unclear. We conducted two survey experiments with 4800 respondents in China and the United States. Participants viewed soft propaganda videos in which either an American or a Chinese influencer described their feelings of freedom in China. The results reveal that foreign influencers did not persuade Chinese audiences but Americans evaluated pro-China messages more favorably when delivered by a fellow American. This suggests foreign influencers improve perceptions of authoritarian rule among their co-nationals, but not within such regimes. Our findings show how autocracies can build global support through foreign influencers, which, given heightened geopolitical competition and the emergence of social media as a dominant news source, has implications for democratic resilience.

Evidence from the early months of the COVID-19 pandemic in the U.S. indicated that the virus had vastly different effects across races, with black Americans faring worse on dimensions including illness, hospitalization and death. New data suggests that our understanding of the pandemic’s racial inequities must be revised given the closing of the gap between black and white COVID-related mortality. Initial explanations for inequality in COVID-related outcomes concentrated on static factors—e.g., geography, urbanicity, segregation or age-structures—that are insufficient on their own to explain observed time-varying patterns in inequality. Drawing from a literature suggesting the relevance of political factors in explaining pandemic outcomes, we highlight the importance of political polarization—the partisan divide in pandemic-related policies and beliefs—that varies over time and across geographic units. Specifically, we investigate the role of polarization through two political factors, public opinion and state-level public health policies, using fine-grained data on disparities in public concern over COVID and in state containment/health policies to understand the changing pattern of inequality in mortality. We show that (1) apparent decreases in inequality are driven by increasing total deaths—mostly among white Americans—rather than decreasing mortality among black Americans (2) containment policies are associated with decreasing inequality, likely resulting from lower relative mortality among Blacks (3) as the partisan disparity in Americans who were “unconcerned” about COVID increased, racial inequality in COVID mortality decreased, generating the appearance of greater equality consistent with a “race to the bottom” explanation as overall deaths increased and substantively swamping the effects of containment policies.

In many human-annotated NLP tasks involving ambiguity or subjective judgment, annotator disagreement reflects epistemic uncertainty rather than noise. Soft labeling (SL), which represents annotations as probability distributions rather than majority-vote (MV) labels, preserves this uncertainty and can improve downstream performance. We extend this perspective to LLM-based annotation by formalizing LLM soft labeling as introducing controlled variation in model-generated annotations to approximate the latent variability underlying human annotations. We distinguish two sources of variation: model-induced (e.g., stochastic decoding and model ensembles) and human-approximated (e.g., persona prompting and human-calibrated in-context annotation). Using the Gab Hate and GoEmotions datasets, we show that SL training consistently outperforms MV training under stronger LLM-based annotation strategies. Model ensembles produce the most informative soft-label distributions, achieving the best human–LLM agreement and downstream classification performance. These findings suggest that scalable LLM-based annotation pipelines can model epistemic uncertainty through diverse model-level variation without explicitly simulating human attributes.

Survey data are common and invaluable in social science research for understanding population processes and supporting policymaking and planning. Depending on the nature and scale, survey data sharing comes with privacy risks, and data collectors and agencies are constrained by disclosure permissions, limiting usage across research groups and institutes. Previous methods for synthetic data generation and deidentification may not entirely prevent information disclosures, or they may sacrifice data quality and granularity. Using a large-scale national voter file at both national and state levels, this paper introduces an end-to-end pipeline to streamline synthetic data generation and evaluation for survey researchers. This study selects four generative approaches based on different statistical assumptions: the regression-based Synthpop, the generative deep learning-based CTGAN and TVAE, and the large language model-based REaLDTabFormer, and compares them to the baseline synthetic minority oversampling technique (SMOTE). We consider three key dimensions of evaluation (utility, fidelity, and privacy) to highlight the strengths and weaknesses of each approach, and systematically evaluate across various datasets and training sizes. The results reveal that Synthpop is optimized for general utility (i.e., fidelity), while TVAE excels in downstream applications (i.e., target-specific utility) but compromises on general utility and potentially risks data overfitting. REaLDTabFormer demonstrates a balanced performance in both general and target-specific utility, whereas CTGAN offers the best privacy protection. We recommend that future researchers select a generative method by considering the trade-offs between performance across various evaluation dimensions, training size, data type, and computational infrastructure.

The rise of visual-content-based platforms such as TikTok has transformed media consumption, political discourse, and cultural expression in the United States. Yet TikTok's ownership by the Chinese company ByteDance has sparked widespread concerns over data privacy and national security, culminating in a temporary platform shutdown and looming federal regulation. In response, an unexpected migration occurred: hundreds of thousands of US users downloaded RedNote, another Chinese social media app. Why would users respond to a national security-driven ban by switching to a different Chinese platform? We investigate this puzzle using a preregistered survey of approximately 1,200 active TikTok users in March 2025—conducted during the 75-day regulatory extension following TikTok's temporary shutdown. Our findings suggest that digital migration was shaped by competing threat perceptions. Perceived threats to digital freedom were positively associated with RedNote engagement, while perceived threats from China were negatively associated. Additionally, we show that these relationships were moderated by users' level of social media dependency and their prior exposure to social media influencers. These results offer empirical support for psychological reactance theory in a geopolitical context, highlighting how perceived restrictions on autonomy can produce symbolic, oppositional behavior. Our study provides the first systematic evidence on digital migration under state-imposed platform regulation and demonstrates how political identity and international conflict shape individual responses to technology governance.

How do portrayals of foreign nations as threats shape public opinion on foreign policy and domestic racial attitudes? This study examines how portrayals of China as a threat influence Americans’ support for U.S. policy toward China and attitudes toward Asian Americans. A content analysis of CNN and Fox News transcripts from 2010 to 2020 shows that both outlets increasingly portrayed China as a threat, although Fox News more often emphasized threats to the United States, whereas CNN focused more on threats affecting other countries, international institutions, and populations. A national survey and a preregistered survey experiment further demonstrate that exposure to China threat narratives, regardless of whether the threat targets the United States or another country, increases support for hawkish U.S. foreign policy and heightens anti-Asian resentment, especially among Republicans. These findings show how portrayals of foreign threats shape public opinion and spill over into racial attitudes at home.

In recent years, the rise of social media has flattened the world’s information landscape, with individuals from all corners of the world stepping into roles of information dissemination and influence previously held by traditional media outlets. This democratization of communication extends to the dissemination of propaganda, where foreign influencers have emerged as critical actors in shaping global perceptions of authoritarian regimes. This paper examines the role of YouTube influencers who promote narratives favorable to the Chinese government in shaping public opinion about China. Drawing on more than 3,000 YouTube videos and 1.8 million comments, we analyze the content of videos produced by these influencers and the engagement that their content receives from the public. We find that these influencers target their content, focusing on political topics toward non-Chinese audiences and on cultural content towards Chinese-speaking audiences. We also find that videos containing explicit ideological narratives receive significantly higher levels of audience engagement. These findings highlight the growing role of foreign influencers in extending authoritarian influence through social media.