This dissertation explores the mechanisms through which authoritarian regimes shape public opinion. It specifically explores the effects of media frames on attitudes towards economic inequality, the influence of victimization narratives on national identity and policy preferences, and the lasting impact of political upheavals on political interest. The first study delves into the role of state-controlled media in shaping public attitudes towards economic inequality. Contrary to expectations, negative framing of foreign inequality and positive framing of domestic inequality reduction efforts do not significantly influence attitudes. Moreover, charitable donations from the wealthy are more effective than property taxes in reducing the perceived severity of inequality and improving public sentiment towards the government among low-income groups, highlighting the greater symbolic impact of voluntary giving compared to compulsory taxation. The second study scrutinizes how the state’s construction of collective memory, specifically through the victimization narrative of historical events, influences national identity, foreign policy preferences, and political support. Highlighting the Boxer Rebellion as an example, it reveals how this narrative strengthens the victim side of the national identity, reduces trust towards foreign nations, bolsters support for hawkish foreign policies, and fosters support for the government, particularly among Chinese citizens without college degrees. The third and last study analyzes the lasting impact of the Maoist era’s societal upheaval, revealing a distinct generation characterized by heightened political interest. This finding, unique to China, demonstrates the long-term impact of systemic state-led persecution, mass mobilization, and a fear-saturated political climate. Together, these studies shed light on the mechanisms through which authoritarian regimes, like China, can shape public perceptions. The research underscores the interplay between state-controlled narratives, media framing, and public opinion, revealing significant implications for domestic stability and international relations.

Trained on massive amounts of data, Artificial Intelligence (AI) has achieved amazing results in generating realistic texts, images, videos, and more. However, the data that gives AI its capabilities - data produced by internet users, institutions, and organizations across the world - inevitably contains the biases and politics that many have hoped AI could overcome. This dissertation studies the way one kind of politics - authoritarian politics - gets embedded in AI's training data and its downstream implications. It seeks to answer the following questions: 1) How do authoritarian institutions shape the behavior and performance of AI? 2) What is the effect of AI on authoritarian politics? 3) how does the interaction between authoritarian politics and AI spill over into democracies?

Using novel experiments on commercial AI systems, open-source large language models, and open-source training data, this dissertation presents both theoretical and empirical evidence on how authoritarian politics serve to enhance but also constrain the utility of AI for repressive ends. Furthermore, it shows how authoritarian politics can spill over into democracies in subtle ways through AI.

My dissertation employs computational methods to explore the complex relationship between media and politics in modern China. It consists of three interconnected studies. First, I analyze how the government uses newspapers to boost its legitimacy, focusing on the publicity of punitive actions against corrupt local officials. Second, I examine intra-party dynamics, revealing how media narratives on corruption are used to weaken rival factions within the Chinese Communist Party. Lastly, I explore the state's control over social media, specifically the implications of geographic tagging in online discourse on public matters. Overall, my work offers a data-driven insight into the intricate interplay of media, political actors, and societal norms in China.

Despite global advocacy efforts, journalism remains a dangerous profession in many parts of the world. While independent media sometimes manage to operate within authoritarian systems, this coexistence is often precarious. Autocrats may tolerate limited press freedom, but only so long as it does not threaten their grip on power. To control the flow of information and minimize public backlash, many modern autocracies rely on selective, surgical forms of media repression. This dissertation investigates the strategies, consequences, and public perceptions of such targeted repression, with a focus on contemporary Eurasia.

This project draws on three original sources of data: (1) a comprehensive corpus of over 1.3 million articles published by twelve independent Russian media outlets between 2011 and 2022; (2) a detailed database of more than 1,000 repressive incidents reported by media professionals in Russia during the same period; and (3) novel experimental survey data collected in Kazakhstan. Using these data, the dissertation brings together three empirical papers to investigate how authoritarian governments do more than just punish dissent -- they also work to manage the visibility of repression, reduce the likelihood of public backlash, and shape how journalists operate over time.

The first paper theorizes and documents how regimes may delay repression in response to journalistic transgressions. Using hand-labeled articles, natural language processing (NLP), and large language models (LLMs), it shows that while publishing sensitive content increases the risk of repression, the timing of these publications does not predict the timing of state retaliation. These findings support the idea that governments may deliberately postpone repression to obscure political motives and maintain plausible deniability.

The second paper turns to public perception, using a survey experiment to test how the timing and framing of repression shapes citizens’ reactions. Participants were exposed to scenarios in which a journalist faced state punishment after publishing critical content, with the timing and framing of the response experimentally varied. Results suggest that disguised repression reduces respondents’ ability to recognize it as politically motivated and weakens expressions of solidarity with the targeted journalist -- highlighting how subtle repression may mute public dissent.

The third paper investigates how repression alters journalistic output. By combining byline-level publication records with data on reported attacks, it evaluates how both direct repression and repression observed within professional networks affect journalists’ productivity, the sensitivity of their reporting, and the tone and specificity of their coverage. The results show that both types of repression lead to a decline in article output and sensitive content, pointing to self-censorship and adaptation in response to risk. The analysis also reveals that journalists embedded in more collaborative networks may experience greater disruption after repression -- suggesting that while networks can offer support, they can also amplify the chilling effects of targeted attacks.

Together, the three papers contribute to the study of authoritarian communication, political repression, and media control by shedding light on how autocrats manage the flow of information. The dissertation blends theory-building with a multi-method empirical approach, drawing from political science, communication studies, and computational tools. Beyond its academic contributions, the findings offer practical insights for those working in media development, policymaking, and advocacy in repressive settings.

My dissertation explores the political logic of ambiguity in Chinese national statutes. Departing from the prevailing view of the delegation literature that treats the amount of statutory ambiguity as essentially a control problem, my dissertation offers an alternative account by considering bureaucratic struggle over policy. The core argument is that ambiguity in law allows Chinese top leaders to navigate and satisfy competing interests of bureaucratic stakeholders. For the regime leader, ambiguity helps facilitate compromise among conflicting actors and overcome legislative gridlock. For the competing stakeholders, ambiguity avoids locking in hostile rules and creates bargaining and interpretative space in the post-legislative stage. I further argue that statute ambiguity is a double-edged sword. It helps facilitate timely bill passage and maintain elite loyalty but runs the risk of reinforcing bureaucratic fragmentation and undermining regulatory coherence. I evaluate these ideas through a combination of qualitative case study of the Anti-Monopoly Law and statistical analyses of large collections of laws, implementing regulations and rules between 1993 to 2021 in China. Using process-tracing and novel measures of statute ambiguity, I find that bureaucratic division over policy encourages both jurisdictional and substantive ambiguities in final law. I also find that jurisdictional ambiguity in law is associated with delay of administrative regulations (executive decree) and fragmentation of departmental rules (ministerial decree). My dissertation contributes to advancing our knowledge of how elite conflict is managed in an authoritarian legislature and how policy power is shared among regime insiders holding divergent preferences. It also reveals the legal source of China’s bureaucratically fragmented system.

Although social media is often optimized for broadcasting visual information, studies of political social media have primarily focused on textual content. Concerns about synthetic images and influence operations have only widened our knowledge gap. This dissertation, comprising four papers, examines the supply of vision-based misinformation generated by autocratic actors, the demand for visual content in democracies across demographic groups, and the extent to which images can shape political behaviors, such as credibility perceptions toward news articles. Additionally, I investigate how information control in autocracies loses its effectiveness during crises. To answer these questions, I combine large-scale digital trace data, open-source pretrained models for computer vision and natural language processing, and causal inference methods. Overall, my work offers data-driven insights into the intersection of visual media, political communication, and information control.

Trying to study environmental politics is like trying to juggle flaming bowling pins while riding a unicycle–it requires the acrobat to be good enough at both juggling (political science) and unicycle riding (environmental science) that they do not immediately crash and burn. My attempt to do so is below. In my dissertation I take three different approaches to answering questions which can contribute to our understanding of the interface between the political forces and the environment. In the first I explore electoral deforestation cycles, where deforestation rates are higher surrounding elections in young democracies.These cycles are most pronounced when the elections are highly competitive, occur in young or weakly institutionalized democracies, and are held in majoritarian systems where politicians can effectively target voters with geographic policies. Here, a set of rules designed to expand political power have the unintended consequence of increased environmental destruction. In the second paper we develop a method for discovering and testing influential concepts and phrases in text. We adapt a neural network with recurrent and convolutional layers designed to make the network’s decisions more interpretable to a different task—to identify phrases and concepts which are highly persuasive to a reader. We then apply this to climate change communication to try to uncover some of the most persuasive concepts both for and against climate change mitigation. In it I evaluate the effects of receiving a formal land title on the behavior of plot owners—usually smallholder farmers—using satellite imagery and machine learning. This case speaks to a substantive question that drives millions of dollars in aid annually—how can we reduce barriers to increasing productivity for the world’s poorest and least food secure regions. It also demonstrates new methods which use existing data to more effectively evaluate future (or past!) interventions. Specifically, I evaluate whether having the boundaries of one’s plot officially demarcated and recorded and the ability to obtain a legal title increase the probability that part of the plot will be converted from annual to perennial crops, whether there will be cropland expansion in the plot, and whether the built-up area in these plots increases. I also test whether land titling results in land-sparing in surrounding areas by increasing productivity on the intensive margin. To do this I develop a set of methods for using satellite imagery to measure changes in land cover based on sub-annual variations in surface reflectance of different wavelengths of light. This allows me to observe outcomes at an annual scale and detect the proportion of plots under different types of landcover. Evidence from a pilot area of analysis shows increased conversion from natural forest to cropland as a result of land formalization.

In an age when Communist ideology has largely retreated from the daily lives of ordinary people, the bulk of the CCP's public communication is still heavily coated in ideological jargons. My dissertation argues that these jargons—which I call "catchphrases"—form a system of public intra-elite communication that is inaccessible to the general public, who lack the necessary skills to decipher the political messages embedded in them. Such a system of public communication through catchphrases helps increase the transparency necessary for elite power-sharing in a closed authoritarian regime. Catchphrases signify ideological and policy positions that cannot be easily revealed through literal interpretation. In particular, significant changes in ideological and policy positions are often signaled by extremely minute and seemingly trivial changes in the wording of catchphrases. In this dissertation, I identify a comprehensive list of 25,776 catchphrases from a corpus containing 2.5 million ideological texts from 1921 to 2019, using an automated method I developed based on strings matching. Going beyond a few legacy-defining catchphrases, this list shows how catchphrases form the "staples" of daily CCP elite communications, offering a "menu of options" representing varying ideological and policy positions that elites can use to reveal their differences in public without appearing divided to the general public. Next, using word embeddings, I measure how seemingly trivial changes in the wording of five pairs of catchphrases shift their ideological and policy positions along three predefined scales: a state-market scale that reflects positions on the party-state's role in the economy, a concentration-collective scale that pertains to positions on the power concentration within the party, and an authoritarian-democratic scale that reflects positions on openness to democratic values and social diversity. Finally, I use survey evidence to show how these minor modifications in wording are noticed by their intended audience, i.e., the political elite, rather than the general public. Moreover, such sensitivity to catchphrases cannot simply be acquired by the general public through training. This allows elites to communicate sensitive political information in public to selective audience and to ensure the controlled release of ongoing and unresolved elite conflicts to wider circles of political power.

The Internet has fundamentally changed society and the way people communicate with one another. Advances in technology have made it possible for virtually anyone to create and distribute information. Social media has given voice to the formerly marginalized, allowing them to express themselves to populations worldwide at the click or tap of a button. Authoritarian governments now find themselves in a ``many-to-many'' communication model of the world, where the government no longer holds a monopoly on information.

I use the examples of prefectural-level police accounts to show in this dissertation how the Chinese government is adapting to this new state of the world. Since it must compete with the plethora of voices on social media, it has learned to mimic the styles and adopt the language of social media users. Over time, police accounts have allocated more of their posting volume to non-police topics and have increasingly applied softer styles to their posts. This strategy appears to be paying dividends for the police, as the posts that have softer styles and are not about police topics tend to get forwarded more often.

The changes wrought by the Internet are not limited to just empowering more voices; the anonymity of the Internet enables the regime's second strategy: buying accounts to interact with its posts, artificially inflating the popularity of the post. Some accounts exert great effort to hide the fact that they are ``fake'' accounts, making it difficult for the outside observer to ascertain the true popularity of a post. The government can also rely on other government accounts to promote its posts.

The other appeal of purchasing forwards and likes is that it is relatively cheap. Many local governments are resource-constrained and lack the funds -- or perhaps even the inclination -- to improve the administration of their social media accounts. Compared to the costs of hiring a social media account manager or training an existing worker, just buying interactions may be much more cost-effective in the eyes of a local government.