Research

Civil Society & Civic Space

Press Freedom and Reporting on Corruption: Evidence from the Machine Learning for Peace Project

We analyze significant declines in press freedom across 62 aid-receiving countries using the RSF Press Freedom Index. Of the 46 major declines we identified, we were able to determine that 15 of which were linked to legal changes and 4 to regulatory changes, allowing us to examine how corruption reporting shifts in response to these restrictions. We then assess the effects of legal and regulatory changes on corruption reporting using data from the Machine Learning for Peace project—110 million newspaper articles across all 62 countries. Our findings reveal that legal changes aimed at intimidating the press do not result in significant short-term reductions in corruption coverage. However, regulatory changes, such as alterations to media oversight frameworks, correlate with immediate declines in reporting. Moreover, regulatory restrictions often follow heightened media attention on government corruption, suggesting that regimes use such measures as rapid responses to increased scrutiny.

Policy Report

Forecasting High-Level Travel Advisories with Machine Learning for Peace Data

DOS engaged with the Machine Learning for Peace (MLP) team to test whether MLP’s high-frequency data tracking domestic political conditions can forecast the onset of high-level travel advisories. Initial results are very promising. We can forecast the onset of serious travel warnings three and six months in the future with a high degree of accuracy. Furthermore, our models provide useful information about the political conditions on the ground that are predicting of future travel advisories.

Policy Report

High Frequency Tracking of Civic Space Utilizing Domestic News Scraping and Large Language Model Classification

Changes in politics and society – like protest movements, waves of arrests, corruption scandals, and legal changes- often happen quickly. Unfortunately, many of the standard indicators that researchers have to track civic space or regime characteristics are measured annually. In recent years, several big data approaches to measuring features of domestic and international politics at higher frequency have emerged, including GDELT and POLECAT. We present some of the advantages of tracking civic space using the Machine Learning for Peace (MLP) infrastructure. We apply customized scrapers to a curated sample of national news sources to provide a monthly measures of civic space activity. To process these articles, we apply large language model classification to track 20 different civic space event types. In this report, we provide a clear use case by comparing national and international news coverage of civic space events in 62 countries, highlighting the shortcomings of data sources that rely solely on international coverage.

Policy Report

Elections as Windows of Opportunity? Civic Space in Democratic and Backsliding Regimes

This study investigates how backsliding regimes and consolidated democracies differ in their treatment of civic space during electoral and non-electoral periods, using high-frequency data from the Machine Learning for Peace project. Analyzing seven aspects of civic space, including censorship, arrests, and protests, the research reveals that backsliding regimes engage in significantly more censorship and arrests during non-electoral periods and intensify censorship efforts in the six months preceding elections. Surprisingly, few significant differences exist between regime types during the actual election month, suggesting strategic behavior by aspiring autocrats. Electorally vulnerable backsliding regimes exhibit distinct patterns, using fewer legal actions and experiencing more significant post-election protest reductions. Additionally, these more vulnerable regimes make fewer arrests pre-election but employ more violence and less censorship during the election month compared to their more secure counterparts. These findings indicate that democratic backsliding is more pronounced between elections, emphasizing the need for sustained attention to governance practices during non-electoral periods to safeguard democratic institutions.

Policy Report

The Effect of Government Intervention on the Operational Decisions of NGOs: Evidence from a Survey Experiment in Three Electoral Autocracies

Non-governmental organizations (NGOs) play dual roles, providing services that improve citizen well-being and engaging in anti-incumbent mobilization. Governments respond with a mix of accommodating and coercive tools, yet the impact on NGO behavior remains underexplored. Using a conjoint experiment on directors of 425 NGOs in Cambodia, Uganda, and Serbia, we measure how government actions impact where and how NGOs prefer to work and which actors they prefer to engage with. We find that government repression and cooptation can restrict and isolate NGOs, but also fuel civic action. NGOs prefer to avoid communities where local governments deploy more repression, cooptation, and anti-NGO rhetoric. We also find that coercion reduces NGO collaboration and public involvement but amplifies public mobilization. These findings have important implications for NGOs navigating governmental constraints, shaping political landscapes, and fostering resilience amidst challenges.

Policy Report

Propaganda and Political Scandals: Evidence from El Salvador

Propaganda contributes to the stability of authoritarian regimes by helping convince citizens of the capacity and competence of the leader. Successful authoritarian leaders use propaganda outlets to manipulate information in at least three ways: to selectively attribute positive news to the regime and negative news to other actors, to distract from negative news and scandals by covering different stories, and to manipulate the reference point of negative news to make it appear less bad. We present evidence of Salvadoran President Bukele’s use of propaganda outlets. We classified the partisan slant of 9,098 newspaper articles published by El Salvador’s three most important mainstream newspapers and the most prominent propaganda outlet, one month before and one month after reporting on a major scandal. Our findings suggest that the response of Bukele’s propaganda outlet to media scandals consists on increasing pro-government coverage to selectively attribute positive news to Bukele and his government in an attempt to distract citizens from the scandal.

Policy Report

#Stoprussia: Weaponizing Social Media for Foreign Support

We provide original insight into how governments solicit foreign support during wartime by examining how Ukrainian officials have used social media during its ongoing war with Russia. We combine the Twitter histories of more than 100 top Ukrainian officials with tens of millions of tweets about the war to identify key networks of foreign consumers that the Ukrainian government targets with messages. Our analysis provides three conclusions: First, Twitter is a strategic tool for communicating with international audiences, as both the volume of government tweets increases dramatically and shifts toward English-language posting as soon as the invasion begins. Second, most of those tweets focus on several different international themes and circulate in 9 distinct international networks, which range from governments to human rights organizations to mass media. Third, The content and sentiment of government messaging impact the extent to which they disseminate across different foreign constituencies. For instance, US and Canadian government twitter networks are much more responsive to tweets about human rights violations than they are to those about security considerations. All told, we provide the first large-scale, systematic test of the efficacy of social media outreach efforts to solicit foreign support during war.

Policy Report

Information, Civil Society and Corruption: Evidence from a Randomized Control Trial in Guatemala

Over the past 10 years, corruption scandals have rattled civic space across many countries. Much programming has focused on disseminating information to citizens about government malfeasance as a means of promoting accountability, but the results have varied considerably. Instead of relying on a complex chain of accountability running from voters to politicians, we engage directly with CSOs and bureaucrats themselves. We test whether delivering information on corruption directly to either the CSOs mobilizing civil society or the bureaucrats involved in procurement provides a more direct means of reducing corruption. Our preliminary findings suggest that information about corruption might lead local civil society leaders to plan more organizing to fight corruption, but only when that information shows that corruption is worse than expected. Similarly, information on corruption leads contracting bureaucrats to reduce the extent to which they prioritize legal documents, firm attributes, and a firm’s past experience when assessing bids for government contracts. All told, our preliminary results suggest an interesting, more direct, path for impacting corruption.

Policy Report

Violence Against Journalists and Reporting on Civic Space

This report analyzes the impact of over 200 arrests and murders of journalists on press willingness to report on politically sensitive topics in the following weeks and months. We combine dated arrest and murder data from the Committee to Protect Journalists (CPJ) with data from the Machine Learning for Peace (MLP) project to examine if the ‘shock’ of a violent event changes reporting on 7 different civic space event types in the weeks and months following the event. We report 3 key findings. First, that coverage of several types of civic space events does not change after violence against journalists. Nevertheless, we find that the arrest of journalists represses reporting on several sensitive topics, including police raids, corruption and defamation cases. In contrast, we find that the murder of journalists increases reporting on corruption, legal action and raids.

Policy Report

The Impact of Legal Repression on Citizen Online Behavior: Evidence from Tanzania’s Jamii Forums

Many aspiring autocrats have passed legal restrictions on citizens’ online behavior. Unfortunately, there is little rigorous evidence on the impact of that legal repression. We provide such evidence from Tanzania, where we analyze the impact of the 2015 Cybercrimes Law on posts on political threads in Jamii Forums — a widely used, citizen-driven online platform in the country. Our analysis of more than 11 million individual posts reveals several insightful findings on the ways in which online media repression succeeds and fails to impact citizen postings.

Policy Brief

Policy Report

High-Frequency Evidence on Corruption in 53 Countries: New Data from the MLP Project

Despite the centrality of corruption scandals to civic space dynamics in many countries, our capacity to understand when, why and where corruption elicits civic responses is sharply limited by constraints inherent in standard annual corruption data. In this report we introduce a new big data approach to measuring corruption that allows researchers and analysts to ask new questions. By measuring the share of monthly news reporting on corruption in 53 countries over 11 years, we provide data on its salience. This measure does a good job of identifying corruption scandals and provides a tool for monitoring corruption in near real-time.

Policy Brief

Policy Report

Political Conflict and COVID-19: Evidence from the Machine Learning for Peace Dataset

In March 2020, the World Health Organization officially declared COVID-19 a pandemic, sparking a global wave of emergency measures designed to combat the virus’s spread. The onset of the pandemic and the perceived need for temporary but dramatic restrictions on fundamental liberties — such as the freedoms of movement and assembly — to preserve public health, shaped political conflict in the ensuing months in important ways. We investigate the impact of COVID-19 on political conflict using high-frequency data on government declarations of emergency, levels of civic activism, and government coercion in the months before and after the onset of the pandemic in March 2020. We discuss several specific cases (Serbia, Albania, Belarus, Ukraine, Ethiopia, and Sri Lanka) to tease out the sequencing of civic activism and government coercion after the onset of the pandemic. We find that most countries saw a sustained decline in government coercion, likely due to voluntary compliance with emergency measures, although levels generally returned to normal within one year. Most countries also saw a brief reduction in civic activity, but levels returned to normal within six months in most countries and within one year in nearly all. The return of civic activity was usually related to elections or instances of government repression, while mobilization motivated by dissatisfaction with COVID response itself was relatively rare. These findings have implications for our understanding of how crises affect political mobilization and conflict.

Policy Brief

Policy Report

The Impact of Legal Restrictions on the Content and Sentiment of Media Coverage in Tanzania

One crucial feature of the ongoing global wave of democratic backsliding is that aspiring autocrats seek to influence the media, oftentimes through legal restrictions on the press and social media. Yet little research has examined how formal and social media respond to those legal restrictions targeting the free flow of information. We develop an original argument linking key characteristics of media sources to the regulatory environment and examine how the content and sentiment of their coverage responds to restrictive media laws. We test our claims using an enormous corpus of electronic media in Tanzania and employ two state-of-the-art neural network models to classify the topics and sentiment of news stories. We then estimate diff-in-diff models exploiting a significant legal change that targeted media houses. We find that critical news sources censor the tone of their coverage, even as they continue to cover the same issues; we also find that international news sources are unable to fill the hole left by a critical domestic press. The paper sheds light on the conditions under which the press can be resilient in the face of legal threats.

Policy Brief

Policy Report

An Early Warning System for Democratic Resilience: Predicting Shocks to Civic Space

Civil society is a powerful force for political change and democratic accountability. Understanding this, a growing number of governments have cultivated a diverse repertoire of repressive tactics, ranging from legal sanctions to outright physical coercion. Advances in big data analytics are endowing governments with new tools, including the ability to anticipate citizen action and engage in preemptive repression. Civil society needs new tools to navigate increasingly sophisticated repression. In this research note, we report on our ability to forecast to civic space using the MLP dataset. Analyzing 9 different civic space event types and 39 countries, we find that ror most country-event pairs, we cannot reliably predict shocks. However, we are able to predict certain shocks in certain places with considerable precision. We accomplish this using interpretable models that reveal the model’s decision-making process. Interpretable models provide a way for practitioners with contextual knowledge to judge how reliable models are in the real world. Thus, we provide the basis for an `early warning system’ that could help civil society strategize around repressive government action. 

Read More

The Effect of Closing Civic Space on Foreign Aid: Evidence from 2.3 Million Donor Projects

How donors respond to closing civic space has important implications for the incentives facing aspiring autocrats intent on democratic backsliding. If NGO laws are effective in cutting-off support for advocacy work and are not met with resistance or repercussions from donors, legal restrictions on civil society are likely to continue to proliferate. We examine how donors respond to such restrictions using data on 2.3 million aid projects, original global data tracking NGOs laws, and a variety of research designs. We find evidence that in response to restrictive NGO laws, advocacy-oriented donors decrease spending on advocacy and maintain spending on development. In short, restrictive NGO laws `work’ from the point of view of repressive governments: donors reduce their support for activities that aspiring autocrats find threatening, such as political advocacy, while funding for service-oriented development projects continues unabated.

Read More

Reporting on Civic Space: Differences in Coverage Between National and International Sources

Getting an accurate picture of any country’s civic space is difficult. While many analysts rely on the international news, the vast majority of news coverage on any given country is the news media in that country. The INSPIRES Machine Learning for Peace team has spent enormous time ensuring it is extracting as much news as possible from national sources. But what are the returns to all that effort?

Read More

Democratic Backsliding and Media Responses to Government Repression

A key feature of the global wave of democratic backsliding is that aspiring autocrats seek to influence the media through legal restrictions. We develop an original argument linking media characteristics to the regulatory environment and test it using a huge corpus of electronic media in Tanzania. We employ two state-of-the-art machine learning models to classify the topics and sentiment of news stories and exploit a significant legal change that targeted media houses. We find that critical news sources censor the tone of their articles but continue to cover the same topics; we also find that international news sources do not fill the hole left by a critical domestic press. The paper sheds light on the conditions under which the press can be resilient in the face of legal threats.

Read More

The Effect of Government Repression on Civil Society: Evidence from Cambodia

To limit oversight by civil society, governments often repress NGOs. However, quantitative research has yet to investigate how restricted civic space impacts the behavior of NGOs operating in diverse sectors. Surveying employees from 106 NGOs in Cambodia, we employ a conjoint experiment to identify how the prevalence of repression affects NGOs’ pursuit of funding via grant applications. We find that although increases in the perceived prevalence of harassment has a stronger deterrent effect on advocacy work, harassment also deters NGOs focused on service delivery. Our results suggest that local officials target both advocacy and service delivery NGOs, but for different reasons.

More Details

Legal Changes & Protest: Evidence from High-Frequency Data

This report uses the INSPIRES data to investigate the association between protests and the passage of laws bearing on civic space. The relationship between protest and legal changes has important policy and academic implications. Practitioners often face decisions about whether to support protest movements in support of legal openings, as well as whether to support movements protesting against legal closures. Previous research provides competing findings on the relationship between protests and legal restrictions on free assembly and civic space. That work, however, has been hamstrung by poor data on both protests and the timing and characteristics of laws bearing on civic space.

Read More

Resurgent Authoritarian Influence

Resurgent Authoritarian Influence: Evidence from the Machine Learning for Peace Dataset

In recent years, scholars and policymakers have expressed growing concern about powerful non-democracies becoming more assertive in their foreign policy. According to popular accounts, this has included increased attempts by Russia and China to influence political outcomes in less powerful nations. In this paper, we introduce a new, high-frequency dataset tracking reporting on influence by Russia and China, describe monthly trends in RAI since 2012, including the specific tools used to influence target countries, and assess empirically whether this resurgence is real, providing evidence that authoritiarian influence has increased dramatically in recent years. In doing so, we provide one of the first tests of an empirical claim driving high-level decision-making in foreign policy and international advocacy.

Policy Brief

Policy Report

The Impact of Resurgent Authoritarian Influence on Civic Space: New High-Frequency Evidence

This research memo reports assess the relationship between Resurgent Authoritarianism Influence (RAI) and changes in civic space. We present evidence that for some countries, increases in RAI activity are associated with near-term changes in civic space. In doing so, we provide one of the first tests of a claim driving high-level decision-making in foreign policy and international advocacy. We find that increases in RAI are more often associated with increasing restrictions on civic space, although increases in RAI are also predictive of decreasing restrictions in some cases. Conversely, civic space events are not predictive of RAI events. Together, these findings provide evidence that influence from Russia and China are not so much responding to civic space dynamics in target countries as they are trying to shape it.

Read More

Resurgent Authoritarian Influence: New Machine-Generated, High-Frequency, Cross-National Data

This research memo reviews the academic and policy literatures on Resurgent Authoritarian Influence (RAI), discusses existing data, and describes the MLP RAI data. We group our 22 RAI events into 5 conceptual categories and summarize the prevalence of reporting on these categories across countries and over time using visualizations, descriptive statistics, and dimensionality reduction. Thhis analysis suggest that RAI activity has been surprisingly consistent over the last ten years, that Russia and China utilize regionally-specific approaches to exerting influence, and that Russia and China deployed similar strategies when dealing with strategically important countries with which they have a strained or hostile relationship.

More Details