Public Governance workshop

The use of Large language models (LLMs) to summarise parliamentary proceedings presents a promising means of increasing the accessibility of democratic processes. However, as these systems increasingly mediate access to political information, filtering and framing content before it reaches users, there are important fairness considerations to address. In this work, we evaluate 5 LLMs (both proprietary and open-weight) in the summarisation of plenary debates from the European Parliament to investigate the representational biases that emerge in this context. We develop an attribution-aware evaluation framework to measure speaker-level inclusion and (mis)representation in debate summaries. Across all models and experiments, we find that speakers are less accurately represented in the final summary on the basis of (i) their position in the speaking-order (speeches in the middle of the debate were systematically excluded), (ii) the language spoken (non-English speakers were less faithfully represented), and (iii) their political affiliations (better outcomes for left-of-centre parties). We further show how biases in these contexts can be decomposed to distinguish inclusion bias (systematic omission) from hallucination bias (systematic misrepresentation), and explore the effect of different mitigation strategies. Prompting strategies do not affect these biases. Instead, we propose a hierarchical summarisation method that decomposes the task into simpler extraction and aggregation steps, which we show significantly improves the positional/speaking-order bias across all models. Critically, we find that language and partisan biases persist, indicating that architectural interventions are unlikely to solve biases that emerge due to bias or imbalance in the training data. These findings underscore the need for domain-sensitive evaluation metrics and ethical oversight in the deployment of LLMs for multilingual democratic applications.

TBA

TBA

TBA

TBA

Public goods are meant to be universal, but they are inherently place-based. This paper systematically measures spatial access to public goods and quantifies the implications of distance to public facilities for income inequality. First, we map all schools and hospitals across Belgium. We compute the distance to facilities for each of the 20,000 neighborhoods and document large spatial inequalities in access to public facilities. Second, we find that this unequal distribution favors high-income neighborhoods: allocating public goods spending proportionally to our access index increases income inequality compared to measures based solely on disposable income. Third, we show that the positive relationship between income and access can be rationalized by a simple model of public goods allocation with an inequality-neutral social planner. Finally, we provide evidence that access is strongly correlated with educational and health outcomes, emphasizing the need to consider the place-based nature of public goods when measuring inequality.

TBA

TBA

TBA

TBA

This paper examines how changes in textbook adoption rules shaped historical narratives in American history textbooks between 1870 and 1930. During this period, most Western and Southern states transitioned from local to state-level textbook lists. We investigate whether it enabled regional interest groups such as United Daughters of the Confederacy to influence how publishers portrayed Civil War and the South. Using a novel dataset of 351 editions of middle-school history textbooks, we employ large language models to systematically measure pro-Southern slant across 75,000 text passages. Leveraging variation in the timing of state adoption laws, we find that state-level consolidation led to 18.5% more pro-Confederate coverage of the Civil War, with effects emerging 5-10 years post-adoption. This increase is partly driven by more positive portrayals of Confederate army leaders which was an explicit long-run objective of the United Daughters of the Confederacy activity. In addition to Southern publishers revising narratives to sell more textbooks in the South, some Northern publishers did so as well, resulting in greater exposure to pro-Southern content in some Western and Midwestern schools. Preliminary analysis using 1936-1945 Gallup survey data suggests that cohorts exposed to these biased textbooks during schooling exhibited more white supremacist attitudes later.

Debates about how politicians handle information typically center on interpretation rather than selection. We focus on this earlier and often hidden step and argue that partisan perspectives shape which evidence politicians attend to in the first place. We test this claim in a real-world field setting during the 2024 Austrian parliamentary election campaign. Three weeks before election day, we emailed 1,822 candidates links to two high-quality research briefings on the same immigration policy reform: one emphasizing how benefit cuts reduce migrant inflows, the other that such cuts increase poverty among immigrants. By tracking click behavior, we measured which type of evidence each candidate selected. Our analysis shows that selection of ideologically aligned evidence is concentrated among extreme parties: candidates from the far-right Freedom Party of Austria disproportionately choose aligned evidence, while centrist candidates show more balanced patterns. A contemporaneous survey of 1,428 citizens indicates that far-right supporters approve of elites' selective attention to evidence, suggesting alignment between elite behavior and mass preferences at the political fringes. These results identify evidence selection as an important stage in policymaking and imply that rising far-right electoral strength may amplify reliance on supportive information, potentially increasing the risk of ill-informed policy decisions.

The forthcoming monograph « The Kurdish Case for Democracy » (Palgrave Macmillan 2026) asks how democratic values and civic culture can take shape without equal rights, and even without a state. Focusing on the Kurdish experience across Iraq, Iran, Syria, and Turkey, it draws on fieldwork, survey research, and historical analysis to explore how people respond to authoritarian rule and long-term statelessness. The first part offers a data-driven look at Kurdish attitudes toward democracy, autonomy, and authority. The second situates these patterns in a broader history of political resistance. Rather than turning to authoritarian or religious extremes, many Kurdish movements have embraced more egalitarian, inclusive models of resistance. The Kurdish Case for Democracy makes the case for bottom-up democratization in the Middle East and challenges the idea that universal values must follow Eurocentric templates. It contributes to wider debates on nationalism, resistance, and the political cultures shaping the region

Who uses revolving doors under democracy and autocracy? Existing research often focuses on one-way, single-time transitions — either from business to politics or vice versa — in either democracy or autocracy, overlooking the revolving door as a dynamic, multi-state process unfolding across regime types over the course of individuals' careers. We propose a typology that distinguishes direct and indirect transitions, and specialized roles from concurrent roles. Using a novel dataset tracking the trajectories of nearly 100,000 public officials and entrepreneurs in Spain between 1920 and 2023, we test whether economic power translates into formal political power, and vice versa, across regimes. We find that regime type shapes transitions from politics to business, but less so from business to politics. Highly connected businesspeople enter politics regardless of regime type. In contrast, although politically influential individuals move to business under both regime types, they may be more likely to do so under autocracy.

How does partisan alignment with a populist leader shape compliance with public health measures? During Mexico’s COVID-19 campaign, the president publicly undermined preventive guidelines even as his administration promoted them. Using precinct-level mobility data and a novel shift-share IV, we show that support for the president’s party increased mobility, infections, and deaths. A calibrated epidemiological model indicates that partisan divergence explains most of the infection gap. The evidence supports a supply-side mechanism in which elite cues—rather than fixed partisan traits—shape compliance during crises. Our findings highlight how populist leaders influence public health through policy, symbolic, and rhetorical authority.

This paper develops a novel Bayesian natural language processing framework to recover the division of labour in inventor teams directly from patent texts. The method estimates the probability that each inventor contributed specific knowledge components, informed by their prior collaborations. Applying this framework to approximately 4 million patents, I validate the approach and introduce a novel measure of individual contributions within teams. The results show that as inventors gain experience, they collaborate on more patents per year. However, within projects, senior colleagues contribute relatively less than their junior collaborators. Importantly, concentrating contributions in the hands of lower-performing team members significantly reduces patent value, highlighting how the structure of collaboration shapes the social returns to innovation.

We investigate the impact of law enforcement via police brutality in protest crackdowns on social capital, political participation, and individual attitudes in a mature democracy. Our study focuses on the protests that occurred during the G8 summit in Genova, Italy, in the summer of 2001. To address this research question, we examine how exposure to police violence in Genova influences changes in social capital, political participation, and individual attitudes. We employ digitized data on the geographical locations and activities of the associations involved in the protest, along with administrative and electoral data, large-scale surveys, and other relevant data sources.

The strategic use of the judiciary against political rivals, often termed lawfare, has historically characterized authoritarian regimes and and is now increasingly a source of concern in democracies. A central empirical challenge is distinguishing uniform enforcement of wrongdoing from the selective targeting of political opponents. Using a regression discontinuity design and newly assembled data from Pakistan, we provide causal evidence on the existence and extent of lawfare. We show that opposition politicians face systematically greater judicial scrutiny, with effects large enough to crowd out potentially legitimate anti-corruption efforts. By contrast, government affiliates facing comparable allegations are far less likely to be pursued, underscoring the asymmetric nature of enforcement. These disparities reshape political competition: convictions of opposition politicians eliminate the well-documented incumbency advantage, weaken challengers, and tilt the playing field toward incumbents. In this way, institutions meant to pursue accountability can be redeployed as instruments of political victimization, distorting both justice and electoral competition. Although democratization is often expected to reinforce rule-based institutions, our findings show that lawfare endures under both democratic and military rule alike, transforming courts from instruments of accountability into tools of political repression.

In this paper, I measure the effect of conflict on the demand for frames of reference, or heuristics that help individuals explain their social and political environment by means of analogy. To do so, I examine how Russia's full-scale invasion of Ukraine in February 2022 reshaped readership of history and social science books in Russia. Combining roughly 4,000 book abstracts retrieved from the online catalogue of Russia’s largest bookstore chain with data on monthly reading patterns of more than 100,000 users of the most popular Russian-language social reading platform, I find that the invasion prompted an abrupt and substantial increase in readership of books that engage with the experience of life under dictatorship and acquiescence to dictatorial crimes, with a predominant focus on Nazi Germany. I interpret my results as evidence that history books, by offering regime-critical frames of reference, may serve as an outlet for expressing dissent in a repressive authoritarian regime.

Why do leaders initiate costly wars? Traditional explanations emphasize empire building or resource extraction. We advance a different view: war as a tool of nation-building. Using the Russian invasion of Ukraine, we show that President Putin’s decision is best understood as an effort to consolidate domestic power by appealing to ethnic Russians and strengthening national identity in the face of declining popularity. Ethnic Russians are systematically more likely to turn out and support Putin, and this effect is not more substantial in places with symbols of imperial history, such as statues of Lenin and WWII monuments. Instead, the mobilization response is concentrated among areas with a pre-existing ethnic Russian presence. These patterns remain robust to alternative strategies that account for electoral fraud. We situate these findings within the broader literature on the causes of conflict and the political economy of nation-building. The evidence is most consistent with war being used less to expand the empire and more to forge internal cohesion, rally support, and preserve regime survival.

Large language models (LLMs) can structure big data that either available or unavailable to researchers. I clearly distinguish two core approaches: retrieval and generation. Retrieval involves leveraging LLMs to extract structured information from extensive document corpora provided by user, while generation taps directly into the model's internal knowledge to produce structured outputs from minimal inputs. I will present papers that use these methods both separately: i) document-based retrieval techniques for extracting key data points and causal relationships from long and messy documents (e.g., www.causal.claims), alongside (ii) generative methods to construct granular production networks (e.g., aipnet.io) and comprehensive keyword dictionaries for political text classification (e.g., www.academicexpression.online). Additionally, I'll cover principles I've encountered regarding the LLM pipeline to effectively scale and adapt these tools to wide-set of research opportunities.

Populations in several countries have become decidedly more polarized in recent decades. Many believe that social media, which facilitates interactions within echo chambers, is partly to blame. These interactions can trigger two distinct effects on the demand for biased news. First, individuals can be influenced by their peers' news consumption, for example, because they value keeping a news diet that is ideologically congruent with that of their peers. Second, individuals might purposefully skew their news consumption in anticipation that their peers will observe these choices. We design a field experiment on Twitter (renamed X in 2023) to separately identify the importance of both mechanisms. Our main result documents that, through these two mechanisms, online interactions with like-minded peers are not a major contributor to the demand for polarized news content. Our experiment induces variation in an individual's perceptions of the political leanings of their peers' news consumption and the visibility of their own news consumption to their social media followers. We track participants' sharing behavior and news consumption, proxied by the news outlets they follow. We find no evidence to support the first channel: our experimental variation influences respondents' beliefs about the news diets of their peers, but they do not respond by changing their own news diets. In contrast, we find that participants alter their news diet considerably when they believe their peers will observe these choices, as in the second channel. Interestingly, individuals primarily wish to present themselves as following a balanced set of news. Therefore, our paper uncovers one mechanism through which social media can attenuate the demand for polarizing content: as these platforms amplify the visibility of user interactions, which increases the importance of social image concerns, users adjust their news consumption to be more balanced.

This study examines the short-, medium-, and long-term impacts of state censorship on knowledge production, focusing on the largest book-banning campaign in Chinese history, initiated during the compilation of the Siku Quanshu (Complete Library in Four Sections) between 1772and 1783. Analyzing publication data from over 161,000 books spanning the 1660s to the 1940s, we find that categories subjected to stricter censorship experienced significant declines in publication during the seven decades following the bans (1770s–1830s). However, beginning in the 1840s, political upheavals and the erosion of state control triggered a resurgence of publications in previously restricted categories. This pattern of suppression and recovery extends to ideas, as evidenced by keyword analysis of book titles. Further analysis highlights the central role of chilling effects in driving these dynamics, with the exit and entry of publishers playing a crucial role in both the suppression and subsequent revival of knowledge production.

We conducted a large-scale field experiment on Facebook to identify behavioral and information frictions in news consumption, and to test interventions aimed at improving the reliability and moderating the slant of users’ news portfolios on the platform. We find that both types of frictions are pervasive. News-following behavior on Facebook is shaped by inertia and passivity in the face of minor “hassle costs” (behavioral frictions). Furthermore, although participants exhibit a broad understanding of the ideological slant of news outlets, they systematically misperceive their reliability (information frictions). To address these barriers, we implement a two-pronged intervention: a user-friendly, platform-integrated interface that enables participants to easily re-optimize the set of news pages they follow, and personalized information about the slant and reliability of both individual outlets and users’ overall portfolios. Our results show that re-optimization alone leads users to moderate their news portfolios on Facebook, while pairing re-optimization with information significantly improves portfolio reliability. These changes move users closer to their self-reported ideal news diets and translate into measurable shifts in off-platform news consumption. Taken together, the findings highlight the potential for simple, scalable interventions to shape online news diets in a way that benefits users and has positive externalities on society.”

Committees often decide on proposals that are inherently multidimensional: e.g., whether a medical drug up for approval is safe and effective, or a new investment’s profitability and risk. In this paper, we first provide a model of multidimensional learning in committees where members may collectively — and strategically — learn about different dimensions during deliberation, and that information is germane to the final decision. We show that the multidimensional nature of deliberations, including the spillovers of information and the preferences across dimensions, affects how committees learn and decide. Then, we use data on rosters, minutes, and the text from transcripts from FDA Advisory Committees, where experts provide advice for the approval of new drugs, biological products and medical devices in the United States, and show that we can statistically identify and consistently estimate this model. Finally, we discuss how alternative institutional designs, including changing committee compositions and information structures, affect policy decisions.

The spread of propaganda, misinformation, and biased narratives from autocratic regimes, especially on social media, is a growing concern in many democracies. Can censorship be an effective tool to curb the spread of such slanted narratives? In this paper, we study the European Union’s ban on Russian state-led news outlets after the 2022 Russian invasion of Ukraine. We analyze 775,616 tweets from 133,276 users on Twitter/X, employing a difference-in-differences strategy. We show that the ban reduced pro-Russian slant among users who had previously directly interacted with banned outlets. The impact is most pronounced among users with the highest pre-ban slant levels. However, this effect was short-lived, with slant returning to its pre-ban levels within two weeks post-enforcement. Additionally, we find a detectable albeit less pronounced indirect effect on users who had not directly interacted with the outlets before the ban. We provide evidence that other suppliers of propaganda may have actively sought to mitigate the ban’s influence by intensifying their activity, effectively counteracting the persistence and reach of the ban.

When are public comments characterized by high information quality more likely to occur in bureaucratic policymaking? We answer this question using a text-as-data approach applied to an original dataset covering more than 20,000 comments across 1,037 policy acts issued by the European Commission. We construct four measures capturing our multi-dimensional concept of information quality of comments. Our argument emphasizes the interplay between the demand and supply of information provision and highlights the critical role of institutional factors in explaining comments’ information quality. We find that high-quality comments are more likely to emerge during the policy formulation stage and, counterintuitively, in relation to EC policy documents that are informationally dense and syntactically complex. We find no systematic co-variation between a policy area’s established or scientific status and comment quality. Our findings provide novel insights into how the design of public commenting procedures and crafting policy acts shape comments’ information quality

We propose a framework for survey design as a market interaction. In our model, respondents choose answer options closest to their true beliefs, forming demand, while crowdsourced contributors supply options under both monetary and non-monetary incentive schemes. This framework yields two theoretically grounded metrics for evaluating the quality of answer sets for multiple-choice questions: (1) consumer surplus, measured as the semantic similarity between the chosen option and the respondent's true beliefs, and (2) market share, defined as the probability of selecting any option from the set. We propose a method for estimating demand that leverages text embeddings, a recent advance in natural language processing. To validate the model, we conduct an experiment and find that the demand model successfully predicts respondent choices. On the supply side, we find that while monetary incentives improve crowdsourced options, supplier characteristics—particularly their own beliefs about the question of interest—play a larger role. Our results suggest that selecting an appropriate crowdsourcing population may be more important than incentive design for obtaining high-quality survey options.

The post-World War I era witnessed a dramatic resurgence of the Ku Klux Klan, with membership swelling to millions across the United States during the 1920s. Drawing on newly digitized military records and KKK membership rolls, we examine the relationship between military service and subsequent Klan participation among white veterans. Reconstructing the WWI draft lottery, we find that white men who were randomly inducted into the Army were significantly more likely to join the KKK than comparable non-drafted men from the same communities. This effect was particularly pronounced among those conscripted from regions with stronger expressions of patriotic sentiment. Our analysis reveals that increased KKK membership was significantly mediated by contact with Black and Jewish soldiers, with the strongest effects observed among white veterans with the most exposure to those groups. These findings contribute to our understanding of how wartime experiences shaped racial attitudes and mobilized white supremacist activity in interwar America. 

The spread of propaganda, misinformation, and biased narratives from autocratic regimes, especially on social media, is a growing concern in many democracies. Can censorship be an effective tool to curb the spread of such slanted narratives? In this paper, we study the European Union’s ban on Russian state-led news outlets after the 2022 Russian invasion of Ukraine. We analyze 775,616 tweets from 133,276 users on Twitter/X, employing a difference-in-differences strategy. We show that the ban reduced pro-Russian slant among users who had previously directly interacted with banned outlets. The impact is most pronounced among users with the highest pre-ban slant levels. However, this effect was short-lived, with slant returning to its pre-ban levels within two weeks post-enforcement. Additionally, we find a detectable albeit less pronounced indirect effect on users who had not directly interacted with the outlets before the ban. We provide evidence that other suppliers of propaganda may have actively sought to mitigate the ban’s influence by intensifying their activity, effectively counteracting the persistence and reach of the ban.

Social media platforms play a crucial role in shaping public discourse and political conflict. But political conflict also shapes behavior on social media. This paper investigates the response of social media users to political events during the height of the independence struggle in 2017 in Catalonia (Spain). The analysis exploits linguistic markers to study retweets within and across Catalan and Castellano language groups in a large Twitter corpus of 26 million tweets during this conflict. To be able to study changes in retweet behavior, the article proposes a statistical decomposition into ‘exposure’—the likelihood of encountering specific content—and ‘retweet rates’—the probability of retweeting content once exposed. Surprisingly, the political events were associated with a dramatic relative increase in retweet rates across language groups and a corresponding explosion of exposure. There is evidence of a realignment, where left-wing content from both language groups was spread more strongly in the other language group.

Many public policies are accompanied by narratives which exist independently of the formal policy text. This paper focuses on narratives around welfare benefits. Popular narratives around welfare benefits have taken many forms, from Reagan's `Welfare Queen' in the US, `welfare Wednesday' in Canada, to `Florida Rolf' in Germany. First, we provide descriptive evidence on the prevalence of narratives about welfare across the UK, USA, Germany and France. Using data from national newspapers across these four countries and natural language processing techniques, we find that on average 1 in 100 articles mentions welfare, with the highest levels in the UK and the lowest in France. Second, focusing on the UK, we look at the causal effect of these narratives on beliefs, welfare benefit take-up and labour market decisions. Using an event study design, around the publication of negative stories, we find a decrease of 4 to 5% in applications to welfare benefits, and a 1 to 2% reduction in job search intensity.

We show that US politicians’ campaign advertisements between 2000 and 2020 are dominated by economic issues such as jobs and taxes, and rarely mention divisive cultural topics such as immigration, abortion, and guns. In contrast, we find that cultural topics dominate political news coverage, particularly cable news. Politicians underperform electorally when they emphasize cultural rather than economic issues, whereas news outlets attract larger audiences when they cover cultural rather than economic topics. We explain these differences in part through a difference in objectives: politicians maximize their vote share, whereas news outlets maximize their audience size. Linking campaign ads with individual voting decisions and high-frequency variation in TV news content with household-level viewership responses, we show that economic content is better for share-maximization and cultural content is better for size-maximization: economic content disproportionately attracts swing voters or swing viewers who would otherwise choose a politician’s or news outlet’s competitor, while cultural content disproportionately mobilizes individuals who would otherwise not vote or not view news. The media’s cultural focus spills over into politics: voters in constituencies exogenously more exposed to cable news weight cultural issues more heavily, and local politicians supply more cultural ads. Estimating a model of political competition and media choice, we show that the entry of cable news channels in the 1990s can explain 40% of the subsequent increase in cultural polarization.

[Study 1] While refugee issues have been heavily politicized in popular host countries, politics is still in flux in countries with little refugee inflows. How do natives respond to refugee inflows in the absence of clear political cues? We examine this question using a case of the unexpected arrival of Yemeni refugees in South Korea in 2018, an ethnically homogeneous country with limited political discourse on refugees. Using supervised text analysis of over 150,000 posts on social media, we show that the arrival of refugees led to a significant rise in engagement and negative attitudes towards refugees without partisan polarization. In the absence of domestic partisan cues, individuals turned to international events, particularly the 2015 European refugee crisis. Hostility toward Yemeni refugees in Korea was expressed with frequent references to the European refugee crisis. These findings broaden our understanding of refugee reception beyond popular host countries and demonstrate how backlash against refugees can spread globally.   [Study 2] How does globalization influence cultural products, and how do consumers respond to cultural changes driven by global influences? This paper uses K-pop, a globally popular Korean music genre, as a case to examine these questions. While the industry has gained economic benefits as one of Korea's fastest-growing export sectors, its success has necessitated internationalizing content to appeal to foreign audiences, making it a prime example of cultural globalization. Analyzing nearly three million YouTube comments (2010–2023) with LLM-aided text analysis, we find rising anti-globalization and anti-foreign sentiments among Korean consumers as K-pop becomes more globalized. These findings illustrate how cultural globalization, despite its economic benefits, can provoke hostility when foreign influence is perceived as diluting national identity, underscoring the importance of cultural dimensions in globalization’s societal impacts.

Political polarization is a growing concern in many countries. Are mass protests merely a sign of increasing cleavages, or do they polarize societies? In this paper, we estimate the impact of Nazi marches in 1932 Hamburg, using granular data from 622 voting precincts during 6 elections. We show propaganda can convince – but it does so the most in areas with high initial support. Importantly, marches can also backfire, repelling voters. Thus, protest matches lead to polarization. These effects diffused through social networks, measured as contagion patterns across neighborhoods from the 1918 Spanish flu outbreak. The electoral effects of social spillovers are of similar importance as direct exposure, and grow over time.

As social media is increasingly popular, we examine the reputational costs of its increased centrality among academics. Analyzing posts of 98,000 scientists on Twitter (2016–2022) reveals substantial and varied political discourse. We assess the impact of such online political expression with online experiments on a representative sample of 3,700 U.S. respondents and 135 journalists who rate vignettes of synthetic academic profiles with varied political affiliations. Politically neutral scientists are viewed as the most credible. Strikingly, on both the ’left’ and ’right’ sides of politically neutral, there is a monotonic penalty for scientists displaying political affiliations: the stronger their posts, the less credible their profile and research are perceived, and the lower the public’s willingness to read their content, especially among oppositely aligned respondents. A survey of 128 scientists shows awareness of this penalty and a consensus on avoiding political expression outside their expertise.

This paper examines how exposure to successful democratic institutions influences subsequent electoral support for constitutional order. It leverages a historical quasi-experimental setting: the 1864 Zemstvo reform, which introduced spatial variation within the Russian Empire in access to democratic institutions. The reform established local self-governments, allowing various social groups, including former serfs, to participate in local elections and elect officials to local assemblies. Additionally, Zemstvo institutions played a key role in expanding public goods provision, such as primary schooling and healthcare. Using population census data, the paper exploits variation across cohorts over time, demonstrating that those exposed to the reform experienced significant gains in human capital and occupational choice. By employing a spatial regression discontinuity design and analyzing the results of the 1917 Constitutional Assembly Elections, the paper shows that the literate population in Zemstvo regions, who benefited from public goods provision, increased electoral support for liberal parties (Constitutional Democrats). The findings highlight the mutually reinforcing relationship between support for liberal democracy and economic progress, which benefits broad segments of the population.

This research examines the historical evolution of corporate purpose and the legal mechanisms affecting corporate charters. Drawing upon UK royal charters and utilizing textual analysis, the study reveals that corporate charters were issued to fulfill some public-oriented purpose. The analysis shows that such conception of corporate purpose was enforced. Employing the case study of the Massachusetts Bay Company, the research demonstrates that charter revocations historically stemmed from corporate activities exceeding specified purposes. Contrary to prevalent contemporary discourse emphasizing corporate purpose as a broad goal, this study highlights the importance of law in ensuring a commitment to a limited, well-defined corporate purpose. Ultimately, this research contributes to a nuanced understanding of corporate purpose, highlighting its historical roots and its implications for contemporary corporate governance and legal practices

Capitalizing on a policy discontinuity created by Stalin's abrupt ban on abortions in 1936, this paper provides evidence of the demographic and public health consequences of restricting abortions using newly digitized archival documents from the Soviet Union. Our difference-in-differences and IV estimates indicate a substantial increase in birth rates post-ban on abortions, aligning with Soviet objectives for population growth. However, the ban on abortions coincided with a substantial increase in infant mortality—a linkage oft insinuated but seldom demonstrated causally. Subsequent scrutiny of recently declassified criminal records from the Soviet Union on illegal abortions indicates that regions with high abortion rates pre-ban show a distinct uptick in illegal abortions and a rise in female deaths directly related to abortions. Female deaths not directly related to abortions are unaffected.

How does local government amalgamation affect public goods provision, economic activity, and nation building? We focus on an administrative reform in Ukraine between 2015-2020, during which smaller local councils (LCs) had the opportunity to voluntarily amalgamate in order to keep a substantially larger portion of their tax revenues and gain greater autonomy over the local administration. By investigating the determinants of the willingness to amalgamate, we show that some of the pre-reform characteristics, such as tax revenues, the share of native Russian speakers, and political preferences, were not substantial predictors of the amalgamation. We first show that the reform positively impacted district level personal income tax collection. Secondly, by employing previously unused data on standardized college entrance exam results, we estimate a staggered difference-in-differences model and show that the reform did not affect the Ukrainian test scores significantly. For math results, it led to a 0.07 standard deviation decrease after four years of exposure. Utilizing the same staggered design, we show that the reform led to 0.06 standard deviations increase in log nightlight intensity per capita after four years of exposure. Finally, using a nationally representative repeated cross-sectional survey, we document the greater self-identification as Ukrainian as opposed to Russian in rural areas.

Laws have increasingly become difficult for the public to understand due to their complexity and the decline in trusted legal journalism. We build a novel AI-powered legal summarizer to produce easy-to-read summaries of the reasoning in judicial opinions. We study the effects of these summaries in three related experiments. First, we show in a survey experiment that, compared to existing expert-written summaries, AI-generated simple summaries of U.S. Supreme Court judicial opinions are more accessible to the public and more easily understood by non-experts. They help respondents understand the key features of a ruling, and have higher perceived quality, especially for respondents with less formal education. Second, we study how these summaries affect policy attitudes and institutional legitimacy in a large-scale survey experiment around the release of the affirmative action decision in June 2023. While the summaries enhance understanding of judicial opinions across issues, they have mixed effects on the acceptance of court decisions. Finally, we explore the effect of these summaries on public discourse in a social media experiment.

Social media has changed the structure of mass communication. In this paper we explore its role in influencing political donations. Using a daily dataset of campaign contributions and Twitter activity for US Members of Congress 2019-2020, we find that attention on Twitter (as measured by likes) is positively correlated with the amount of daily small donations received. However, this is not true for everybody: the impact on campaign donations is highly skewed, indicating very concentrated returns to attention that are in line with a ‘winner-takes-all’ market. Our results are confirmed in a geography-based causal design linking member’s donations across states.

How does social media affect the composition of political protests in the United States? Using early adoption of Twitter at the 2007 South by Southwest (SXSW) festival as a plausibly exogenous source of variation in county-level Twitter penetration (Muller & Schwarz, 2023), and comprehensive data on protest events, this paper finds that Twitter penetration increases the frequency of protests overall, but also radicalizes them. Twitter disproportionately fuels protests with participation of “extreme” groups—groups that are particularly militant, radical, or hateful. These effects do not depend on the topic of the protest nor political lean- ing. I also present survey evidence suggesting that coordination is not the only mechanism driving these results: An increase in county-level Twitter penetration implies an increase in respondent’s willingness to justify violence against other people, normalizing the participation in extreme groups and extreme protests.

This article aims at understanding the policy-making process by examining the relationship between access to policy-makers and lobbying success. I collect unique large-scale textual data on the content of lobbying activities and their subsequent policy changes. I identify instances of lobbying success with two complementary approaches: one based on a plagiarism-detection algorithm and the other on GPT. I match this novel data with meetings held between policy-makers and interest representatives to measure access to policy-makers. It reveals notable disparities in access, with the business sector having more access to policy-makers than the civil society. Moreover, I find that access to policy-makers is associated with a higher likelihood of lobbying success, by 11 percent of one standard deviation. This increased success likelihood is larger for entities with more access, as measured by the number of meetings they have. Distinguishing access to policy-makers contemporaneously or before the discussions on a policy, I find that prior access to policy-makers is also associated with higher chances of success. It suggests that reputation and connection-building play a critical role. These results are driven by the business sector, composed of companies and business associations. It indicates that in addition to having more access to policy-makers and being better politically connected, companies and business associations derive greater benefits from these connections. In contrast, NGOs with access to policy-makers do not display an increased probability of success.

Understanding behavioral aspects of collective decision-making is an important challenge for eco- nomics, and narratives are a crucial group-based mechanism that influences human decision- making. This paper introduces the Character-Role Narrative Framework as a tool to systematically analyze narratives, and applies it to study US climate change policy on Twitter over the 2010- 2021 period. We build on the idea of the so-called drama triangle that suggests, within the context of a topic, the essence of a narrative is captured by its characters in one of three essential roles: hero, villain, and victim. We show how this intuitive framework can be easily integrated into an empirical pipeline and scaled up to large text corpora using supervised machine learning. In our application to US climate change policy narratives, we find strong changes in the frequency of simple and complex character-role narratives over time. Using contagiousness, popularity, and sparking conversation as three distinct dimensions of virality, we show that narratives that are simple, feature human characters and emphasize villains tend to be more viral. Focusing on Donald Trump as an example of a populist leader, we demonstrate that populism is linked to a higher share of such simple, human, and villain-focused narratives.

Do mainstream media outlets actively amplify politicians' social media statements, and does this amplification influence individuals' political opinions? I first study cable news coverage of Donald J. Trump's tweets using novel high-frequency coverage measures for CNN, Fox News, and MSNBC. I show that Trump was able to set the agenda of cable news outlets through Twitter, with cable outlets covering his tweets minutes after these had been posted. I then leverage a large public opinion survey to investigate the impact of TV coverage of Trump's tweets on public opinion. I find that CNN's coverage of Donald J. Trump's tweets caused CNN viewers to decrease their approval of President Trump hours after coverage. Conversely, coverage on Fox News resulted in an increase in President Trump's approval ratings among Fox News viewers. These findings shed light on a new channel through which social media impacts individuals' political opinions, potentially fuelling political polarization.

This paper studies nation-building in a fragmented society. We document the adoption of a common language and the construction of a national identity in France. Using a natural experiment and drawing on a novel dataset on the languages spoken across municipalities on the eve of the twentieth century, we establish that state intervention in the provision of education brought homogenization. To understand why nation-building was successful, we study heterogeneity and find that elites and the demand for education were instrumental in driving assimilation. Finally, we document further impacts on identity and ideology in the twentieth century.

How does the success of a movement influence the most extreme fractions within it? We examine this question in the context of the ascent of PEGIDA, Germany's most prominent right-wing movement since World War II. We combine a difference in differences strategy with variation in local weather conditions to show that municipalities experience a surge in hate crimes following protest days with a larger number of participants. The effects are most pronounced for municipalities with higher social media penetration and a more economically vulnerable refugee population. We also document that large protests have substantial spill-over effects to other municipalities that propagate through social media networks rather than geographic proximity. Our findings suggest that a public display of support for a movement — measured in terms of protest attendance and favorable public perception — can reduce the social cost of radical behavior and embolden extremist factions within the movement.

How does the advent of new information shape societal norms and, consequently, behavior? We delve into the aftermath of Bolsonaro's triumph in the 2018 Brazilian presidential election, examining its impact on the prevalence of online hate speech. Leveraging Twitter data spanning 2017 to 2019, we employ Natural Language Processing (NLP) techniques to distinguish tweets containing hate speech. To precisely trace the impact of Bolsonaro's election in online expressions of hate, we adopt a difference-in-differences methodology, utilizing the election outcome as an information shock. We observed a significant surge in instances of hate speech via Twitter after the elections. This phenomenon was particularly pronounced in municipalities where Bolsonaro's support was comparatively low. These results find reinforcement in the individual-level analysis, indicating that intensive and extensive margins of individual hate speech contributed to the overall increase. We interpret these findings through the lens of a belief updating mechanism, specifically emphasizing the process of revising social norms dictating what is deemed acceptable to say (or not) in public.

This paper investigates the political-economic determinants of nuclear energy investment using the Chernobyl accident as a natural experiment. We document several facts. First, Chernobyl reduced worldwide growth in the number of nuclear power plants. Second, the reduction is driven by increased construction delays in democracies and results in the prolonged use of older and less safe plants. Third, the nuclear growth slowdown in democracies is more pronounced in the presence of large fossil fuel reserves. Finally, nuclear plant capacity reduces coal-fired power plants and air pollution. The results for the UK parliament and the US Congress provide suggestive evidence that lobbying by energy groups was an important driver of the reduction in nuclear energy investment in democracies and, in the long run, likely undermined efforts to increase nuclear energy safety and moderate climate change.

We examine the ownership structure and reach of thousands of news websites in the U.S., Canada, and Europe. By combining data on domain registrants, firm ownership, and domain-level web traffic, we draw detailed ownership networks behind news domains. We find that more than half of these websites have one ultimate owner. Otherwise, they tend to have highly diversified ownership structures, making it non-transparent as to who is ultimately responsible for the content. Moreover, the market concentration for online news is moderate to high and varies substantially across countries. Financial industry firms ultimately own substantial shares in the online news market.

We study the causal impact of a large and unexpected return migration episode on political outcomes during a democratic transition that was characterized by severe political and economic instability. We analyse how the forced displacement of close to half a million settlers from Portuguese speaking African countries, motivated by the eruption of civil wars in these territories, influenced election outcomes after the 1974 Carnation Revolution in Portugal. We apply a difference-in-differences with a continuous treatment assignment based on the share of repatriates per municipality. To deal with potential endogeneity issues, we instrument this with two shift-share variables that rely on very detailed census data covering the universe of repatriates and including information on their regions of birth. We find that repatriates significantly increased voting for right-wing parties in the ten years after the Revolution.

We study how remembrance of an authoritarian regime impacts privacy concerns. Our main hypothesis is that Germany's culture of Holocaust remembrance (Erinnerungskultur) focuses Germans’ attention on the risks associated with private data ending up in the wrong hands. One example of this culture of remembrance are the Stolpersteine, plaques on the sidewalk signalling that a victim of Nazi persecution lived at a given address. We first use a detailed street-level imagery dataset of Berlin to relate the location of the Stolpersteine to a novel geolocated measure of privacy concerns: whether a person asks for their building to be blurred on a street-level imagery provider. We show that there exists a positive relationship between the amount of Stolpersteine near a person’s house or workplace, and the probability that this person will ask the imagery provider to blur the front of their house. This relationship is very localized, as most of the effect concentrates on Stolpersteine that are less than 10 meters away. We show through an experiment that when Germans are primed to think about the Stolpersteine and the Nazi Era, they respond by reading the experiment's consent form more thoroughly, which we take as another alternative measure of privacy concerns.

This paper develops a model of lobbying over industrialization policies and tests its predictions against the data on public petitions to the British Parliament and the US Congress in the 18th-19th centuries. Our theory integrates endogenous lobbying over industrial policies into the standard two-sector model of structural change, and predicts that the intensity of such lobbying first increases at the early stages of structural change, and then declines when the traditional sector contracts. This prediction finds support in the data on industrialization-related petitions in both Britain and the US. Moreover, the model predicts that places with historically more concentrated capital ownership are more successful in lobbying for industrialization. The opposite holds for the concentration of land ownership. We find support for these predictions in (i) data on petitions, (ii) counterfactual simulations, and (iii) historical examples from the 19th century Prussia and from the Middle East.

We study the impact of a change in the content of religious practice on fertility decisions. We do so in the context of the Georgian Orthodox Church, where in December 2007, in a move to reverse declining fertility rates, the church’s Patriarch began to personally baptise third- or higher-parity children. At the macro level, using a synthetic control method and interrupted time series analysis, we find suggestive evidence that Georgian fertility rates did rise in response to this intervention. We validate this result using microdata on fertility histories and religion from a representative sample of Georgian women, exploiting variation in religion, ethnicity, marital status, and prior parity to identify women exogenously treated by the patriarch’s campaign. We find that the baptism campaign significantly increased fertility for treated women (i.e. Georgian Orthodox women, especially if married). It also increased marriage and reduced reported abortion. All of these outcomes were desired targets of the intervention. Our results show that even in industrialized, educated, low-fertility societies, traditional authority figures making use of religious discourses may be able to influence fertility patterns.

We study the media discourse and public opinion on climate change in the aftermath of extreme weather events. In both cable news and local media, and for both national-interest and local events, we find that left-leaning media consistently increase their coverage of climate change in the aftermath of natural disasters, while conservative media do not, despite equal disaster-related coverage. We then link the experience of disasters to concerns about climate change and the environment expressed in large-scale electoral surveys. We find a polarizing effect: disaster experience increases environmental concerns among liberal respondents, but has the opposite effect on conservative respondents. Both effects are driven by areas where the ideology of the respondent is in the minority.

Social media companies are under scrutiny for the prevalence of hateful content on their platforms, but there is scarce empirical evidence of the consequences of regulating such content. We study this question in the context of the ``Network Enforcement Act'' (NetzDG) in Germany, which mandates major social media companies to remove hateful posts within 24 hours. Using a difference-in-differences strategy, we find that the law was associated with a statistically significant reduction in toxic posts by far-right social media users. Further, we show that the NetzDG reduced anti-refugee hate crimes in towns with more far-right Facebook users. Together, these findings suggest that online content moderation can curb online hate speech and offline violence.

In this presentation we discuss our work based on a global, sub-national data set on cross-societal transmission. We use data on how words sound, how they are spelled, and what they mean, to estimate the etymological roots of each word and identify loanwords, using machine learning techniques. Societies with more loanword exchange share more societal traits in common, reinforcing the idea that cross-societal linguistic transmission proxies more generally for the diffusion of traits between groups. We use this to show that improved gains from trade causes cultural convergence, but that economic leverage determines the direction of convergence within a pair of language groups. Finally, we introduce a purely language data-driven approach to exploit the relevant information encoded in language to estimate the geographic origins of a historical phenomenon. We apply the methodology by tracing the history of religious words back to their originating languages to estimate the geographic origin of religious spread for each of Buddhism, Hinduism, Islam, Judaism, and Christianity within 500km of historical estimates.

Immigrants are a large and growing unenfranchised group across many developed countries. Does immigrants' enfranchisement affect how politicians respond to immigration? I study the unique UK context, where immigrants from Ireland and the Commonwealth have voting rights in all elections immediately upon arrival, but these rights are not accorded to other immigrants. I analyse how politicians discuss immigration using text analysis of the universe of speeches in the UK parliament and how MPs vote on immigration bills between 1972 and 2011. I use a shift-share instrument exploiting pre-existing settlement patterns to address immigrants' endogenous location choice. I find that politicians exposed to higher enfranchised immigration spend more time in the parliament discussing issues that affect immigrants positively, yet they vote to increase immigration restrictions. Enfranchisement leads to more political engagement of immigrants, and politicians respond to this engagement. The political cost of favouring enfranchised immigrants is compensated by restricting future immigration.

This paper analyses the evolution of French urban areas from a historical perspective. Using building footprints extracted from collections of digitised historical maps covering mainland France in the 18th, 19th and 20th centuries, we define consistently urban areas and analyse their trajectories along the urban hierarchy. Our findings highlight increasing urbanisation with fewer and larger urban areas, consistent with agglomerations economies. Disaggregate analysis reveals significant heterogeneity, with the emergence, persistence, disappearance of urban areas.

How do politicians use metaphors in their speeches? To provide evidence on this question, we apply a deep-learning-based metaphor detection model to a historical corpus of annual State of the State speeches given by U.S. governors, ranging from 1995 to 2022. Across 9 socio-economic topics, we present the following descriptive findings. First, metaphors are most commonly used on fiscal and economic issues. Second, Democratic governors employ more metaphors on environmental issues relative to Republican governors, who in turn express more metaphors on moral values. Third, we confirm that the language used to express political metaphors is emotionally charged, with a degree of heterogeneity. Our emotion scores increase the most in presence of a metaphor on subjects related to the economy, fiscal issues, and moral values.

I study the economic spillovers of convict labor on local labor markets and firms. Using newly collected panel data on U.S. prisons from 1886 to 1940, I calculate each county’s exposure to prisons. I exploit quasi-random variation in a county’s exposure to the capacities of pre-convict-labor prisons as an instrument. I find that competition from cheap prison-made goods led to higher unemployment and reduced wages (particularly for women) in counties that housed competing manufacturing industries. Affected industries innovated away from the competition and thus had higher patenting rates. I also document that technological changes in affected industries were capital-biased.

Redistribution could be deliberately designed to trigger a civil war. How did the Chinese Communist Party (CCP) rally millions of farmers to win in 1949? The crucial step was to initiate land reform through class struggle, empowering farmers to violently grab land from landlords. Farmers desired land ownership but feared reprisals from landlords, who were backed by the Kuomintang (KMT) government. Therefore, farmers had to choose between joining the CCP's army to defend their land and free-riding. Adopting a Difference-in-Difference design and examining the death records of 566,161 communist soldiers, I find that for counties within 82 kilometers of KMT forces, a greater share of land redistribution to farmers encouraged farmers to fight, leading to a rise in CCP soldier deaths after land reform. However, for counties that were farther than 82 kilometers from KMT forces, a greater share of land transfer to farmers discouraged farmers from fighting (free-riding), resulting in fewer soldier deaths after land reform. A model of class struggle for land ownership explains the two different patterns. This paper develops a novel theory of war mobilization and partially explains the emergence of communism in the twentieth century.

We quantify firms' regulatory compliance costs from 2002 to 2014 in terms of their labor spending to adhere to government rules. Detailed establishment-level occupation data, in combination with occupation-specific task information, allow us to recover the share of an establishment's wage bill owing to employees engaged in regulatory compliance. Regulatory costs account on average for 1.34 to 3.33 percent of the total wage bill of a firm, totaling between $79 and $239 billion in 2014. We investigate the returns to scale in regulatory compliance and find an inverted-U shape, with the percentage regulatory spending peaking for a firm size of around 500 employees. Finally, we develop an instrumental variable methodology for decoupling the role of regulatory requirements from that of enforcement in driving firms' compliance costs.

How has mobile internet affected political polarization in the United States? Using Gallup Daily Poll data covering 1,765,114 individuals in 31,499 ZIP codes between 2008 and 2017, I perform a difference-in-differences analysis and an instrumental-variable design to show that, after gaining access to 3G internet, Democratic voters became more liberal in their political views and increased their support for Democratic congressional candidates and policy priorities, while Republican voters shifted in the opposite direction. This increase in polarization largely did not take place among social media users. Instead, following the arrival of 3G, experienced internet and social media users from both parties became more pro-Democratic, whereas less-experienced users became more proRepublican. This divergence is partly driven by differences in news consumption between the two groups: after the arrival of 3G, experienced internet users decreased their consumption of Fox News, increased their consumption of CNN, and increased their political knowledge. Polarization also increased due to a political realignment of voters: wealthy, well-educated people became more liberal; poor, uneducated people—more conservative.

What are the long term effects of protests? This paper studies how the level of attendance at simultaneous marches organized by the 15M (the Spanish Occupy movement) impacted electoral behavior and political attitudes in the following decade. Using regional variation in weather shocks as an instrumental variable for the level of attendance at simultaneous marches, I find that cities with higher attendance are more concerned about corruption and vote more for left wing and anti-corruption parties and less for far-right parties. Using novel data from Twitter, I document, for the first time, a higher uptake of social media platforms after an offline protest and a persistent difference in online activity in cities with higher attendance. Using survey data, I also show a higher and longer-lasting electoral effect for people that have a social media account. Overall, this paper shows that street protests can have long-lasting effects on political concerns and electoral choices, explained, in part, by the creation of a persistent online social network.

By Alessandra Stampi-Bombelli, Gloria Gennaro, Elliott Ash, Dominik Hangartner A constant of human history is the migration of peoples in search of a better future. In destination countries, these new arrivals come into contact with both the host population as well as already established immigrant communities. How does the arrival of new immigrants affect the perception of outgroup distance among the native majority group? And do new arrivals also change the perceived distance between the host population and existing immigrant groups? We address these questions in the context of the Age of Mass Migration (1860-1920), a period during which sizeable and diverse groups of migrants arrived on U.S. shores. Applying advanced computational linguistics techniques to a newly processed corpus of over 1.8 million newspaper issues (9 million pages) published by 3,675 local outlets in that period, we present a novel text-based measure of perceived socio-cultural distance between U.S.-born natives and 32 immigrant groups. For each mention of an immigrant group, we compute a distance measure that captures whether the group's framing more closely resembles contexts used when portraying immigrants, rather than natives. We use this time- and county-varying outcome to analyse the short- and medium-term effects of immigration inflows on local perceptions of socio-cultural distance toward the arriving and existing immigrant groups.

Promissory representation holds that political parties make promises to voters during election campaigns and generally keep those promises after elections if they have the opportunity to do so. Specific campaign promises let voters know where parties stand on issues, and voters' assessments of governing parties' past records of pledge fulfill- ment is one way in which parties are held to account. Despite the centrality of promise keeping to representation, we know little about how it is affected by economic global- ization, which is one of the defining characteristics of the modern world. We argue that globalization reduces governing parties' ability to keep their campaign promises. Inter- national economic integration increases uncertainty about the feasibility of promises, imposes legal constraints in the form of international commitments that may impede promise keeping in unexpected ways, and empowers market actors that lobby govern- ments when promises threaten their interests. We test the empirical implications of our theory with a mixed-methods approach that combines a large-n quantitative com- parative analysis of pledge fulfillment with a typical case study to trace the underlying causal mechanisms of the theory. The findings indicate that international economic integration exerts a large negative effect on the likelihood of pledge fulfillment in a broad range of contexts and that the hypothesized mechanisms are clearly observable in the detailed case study. These findings have important implications for democratic representation in a globalized world.