How International Organizations Change National Media Coverage of Human Rights

Abstract

How do international organizations change the discussion of human rights violations, and how does their message reach the broader public? I show that national media is a key conduit. I analyze media coverage from the Philippines to show that the content of media coverage of the “war on drugs” changed after a major decision by the International Criminal Court. The ICC increased the proportion of media coverage focusing on human rights by triggering contestation between pro- and anti-human-rights actors. Original survey experimental evidence shows that this coverage shift blunts the effect of ICC actions on support for the war on drugs. This highlights an indirect role for international organizations in shaping media coverage and helps explain why they have struggled to win public opinion battles. Though their actions amplify the voices of human rights supporters, media coverage concurrently amplifies the voices of their opponents, which can crowd out coverage that might otherwise have decreased support for problematic polices.

In 2016, Rodrigo Duterte was elected the sixteenth president of the Philippines, after campaigning on a promise to take a tough stance on drugs. Immediately after his election, he delivered by launching a brutal “war on drugs” (WOD), with extrajudicial killings claiming over 12,000 lives by 2018. The WOD continues to be a pressing, ongoing human rights disaster.¹

Particularly striking is that Duterte's administration conducted its WOD directly in the shadow of international law. The Philippines ratified the Rome Statute of the International Criminal Court (ICC) in 2011, which exposes Duterte and other nationals to prosecution for crimes against humanity. The ICC announced a preliminary examination of the WOD in February 2018 and a full investigation in September 2021. Despite the devastating human consequences of the WOD and clear international legal ramifications, the WOD and Duterte, personally, remained very popular among the broader population.²

The stubbornness of public opinion and the ICC's struggle to gain traction in the Philippines are not unique. The recent pushback against international institutions in developed countries, exemplified by Donald Trump's election in the US and Brexit in the UK, is well documented.³ Yet an equally important, different battle has been waged in developing countries, where leaders bristle at the perceived imperialism of Western institutions, especially those focused on human rights. The period from 2016 to the present has seen many setbacks, with ICC cases collapsing after losing public opinion battles in polarized settings, as in Kenya. Five countries have either withdrawn or threatened to withdraw from the Court. Other ICC members have flaunted their obligations to arrest suspects traveling abroad. Prominent academic and policymaker voices have cited these events when questioning the value of these institutions.

What explains the difficulty facing international organizations (IOs) in these public opinion battles? Understanding how an IO's actions can amplify competing voices in the media in the targeted country provides an important part of the answer. This study joins recent work arguing that national media is a crucial conduit through which the messages of IOs and pro-compliance actors actually reach mass audiences.⁴ Media coverage and its content provide the main point of contact for citizens with political events concerning human rights situations like the WOD. Even though public support is critical for the effectiveness of many IOs, the vast majority of citizens will not come in direct contact with an IO or NGO representative. Yet their statements, messages, frames, and arguments are discussed in the media, which can affect public opinion and ultimately policymaker choices.⁵ The media is thus an important link between IO actions, elite politics, and any subsequent effects on mass politics.

Here, I provide systematic evidence for how media coverage of the WOD shifted after the ICC's first major action in the Philippines. The ICC remains almost entirely in the background of the politics of the WOD until the announcement of a preliminary examination. From a research design perspective, this creates a unique opportunity to observe the media discussion of the WOD from before and after an important, external IO action. The Philippines’ political and media environments also share characteristics with many countries where IOs operate. The Philippines is an imperfect democracy with a partially free media ecology that is not completely controlled by the state.

Empirically, I examine a corpus of thousands of documents from Philippine online newspapers. I use machine learning to classify articles based on their relevance to the WOD. I then use topic modeling to assess how qualitative features of coverage changed after the ICC's announcement. The Court's actions are associated with an increase in the proportion of WOD coverage that relates to human rights. After the ICC's examination, a greater proportion of the coverage incorporates topics related to human rights. This increase is not simply a result of coverage of the ICC itself. The additional human rights content pertains to topics distinct from the ICC and its examination.

This relationship between ICC actions and the content of media coverage is most consistent with a theoretical framework that treats IOs as triggers of contestation.⁶ The ICC examination is followed by greater coverage of pro-human-rights actors, who seize the opportunity to amplify their advocacy. But this is accompanied by increased coverage of the retorts and countervailing efforts of actors supporting the WOD. The increased human rights content consists mainly of coverage of public spats between human rights advocates and their political opponents. To the extent that the ICC shifts the conversation toward voices that espouse human rights and respect for international law, it also amplifies their detractors. While the court has amplified certain pro-human-rights voices, media coverage has focused on the back-and-forth, action–reaction cycle of public statements by those supporting and opposing the WOD.

These findings help us better understand the political frictions IOs contend with—beyond the ICC and the Philippines. Ultimately, IOs will enjoy their greatest ability to compel compliance from recalcitrant states when they have public support and can help sway opinion on a particular issue. Amplifying human rights voices helps with that goal. However, to the degree that this goes hand in hand with amplification of contrary voices, this complicates their ability to generate bottom-up pressure for compliance. The IO sparks heated contestation that is reflected back to the population through the media.

Using evidence from a nationally representative survey experiment, I further show how this dynamic can blunt the ultimate effect of an IO's action in opposition to problematic policies like the WOD. Related work in the context of the US Supreme Court suggests that two-sided media coverage—like that found in coverage emphasizing contestation—blunts the effect of legal decisions on public opinion.⁷ I show that a similar dynamic arises from coverage of the ICC. Randomly assigning respondents to read an article about the ICC without contestation significantly lowers support for the WOD. Yet reading a similar article that emphasizes contestation negates much of this effect, restoring support for the WOD. The media content analysis and survey experimental evidence enhance our understanding of why IOs have had such a difficult time using their platforms and prominence to sway popular opinion, even in somewhat open and democratic societies, like the Philippines.

These findings also highlight a key difference between the politicization of IOs in different contexts. For some contexts—like the populist surges associated with Brexit in Europe or the election of Donald Trump in the United States—the recent backlash against IOs can be best understood within a framework of politicization.⁸ There, political entrepreneurs seize on a low-salience issue and politicize it, molding ambivalent public opinion against international cooperation. Yet, in many developing-world (and developed-world) contexts, IOs act where the underlying issue already has high salience. The IO steps into the fray of pitched political contests with well-drawn battle lines.

The future effectiveness of many IOs will require figuring out how to win—or at least not fan the flames of—public opinion battles over the issues they seek to address. Contestation can catch IOs like the ICC flat-footed, as they spend only a small proportion of their budgets on public relations and lack coordinated communication strategies.⁹ They often operate at arm's length deliberately, to avoid appearing political. The last decade has shown that IOs cannot avoid contestation or limit politicization. We are unlikely to see a return to days when IOs and international cooperation fly under the public radar. IOs and their proponents will need to find communication and political strategies that can compete in the open marketplace of ideas, argumentation, and public opinion, as channeled through the media. Since the past is prologue, IOs experiencing increasing levels of politicization and contestation, like the EU or World Trade Organization (WTO), can learn from the lumps taken by the ICC.

International Organizations and Media Content

Theories of IOs and Mass Politics

IOs generally lack direct enforcement power. But they can influence member-state behavior indirectly, through subnational actors. Several overlapping types of theories link an IO's action to reactions from subnational actors.

In one set of theories, an IO acts like an alarm that provides information on government practices. This helps subnational actors, like interest groups or voters, better demand that politicians implement their preferred policies.¹⁰ Specific to the ICC, elsewhere I develop a model in which the court transmits information about the culpability of a government actor.¹¹ This induces pro-compliance actors at the subnational level to increase their efforts to influence government policy, for example, in deciding whether to remove or retain the accused politician. This increases the probability of compliance, but the marginal effect of the IO on compliance is moderated by the response of anti-compliance actors.

A related set of theoretical arguments highlight how IOs alter the social and institutional context in which governments make decisions. In the spiral model, IOs and transnational advocacy networks empower and legitimate subnational groups. Shaming from international actors and pressure “from below” from domestic actors push policymakers toward compliance with existing commitments.¹² Specific to the ICC, Jo and Simmons argue that “when community norms [against crimes and impunity] are challenged … by ICC actions or statements, there is significant potential for a social reaction to law violations.”¹³ Nouwen calls this a “catalytic effect,” where ICC actions spur general public debate over transitional justice.¹⁴

However, few citizens come in direct contact with the actors in these theories.¹⁵ Outside of victim communities, litigants, or activists, IOs rarely interact with citizens. Most international legal institutions intentionally operate at arm's length, for fear of appearing political.¹⁶ If IOs affect mass politics, what forms the connection between the two? And if specific subnational actors are mobilized by an IO, how do their ensuing campaigns reach the public?

The provision of information via the media in the targeted country is a critical intermediate step for these theories. The vast majority of citizens “experience” the events in those theories through media coverage. The media is a complex actor, acting as both a conveyor of information and a reflection of citizen interests.¹⁷ The media delivers direct, factual information about events. It also conveys messages and frames from political actors.¹⁸ Simultaneously, it responds to the demands of citizens, reflecting a marketplace equilibrium between the media, the public, and politicians. Media members choose among actions and frames supplied by political actors, keeping in mind that their readers demand different types of coverage. A country's level of media freedom moderates the degree to which IOs might affect media coverage. The theoretical arguments here apply best to countries without complete state control of the media or domestic politics—features common to many of the places where IOs like the ICC operate.

The Effect of IOs on Media Coverage

How, then, might IOs affect the content of media coverage? By content I mean the angles, frames, and voices emphasized in the coverage of an issue. For example, the US's use of waterboarding was framed as “torture” at some times and “enhanced interrogation” at others, which affected public opinion.¹⁹ Media coverage of LGBT issues in Eastern Europe alternatively highlighted the rights of LGBT citizens and the right to defend the nation against hyperliberal external interference.²⁰ The discursive framing of human rights norms that takes place in the media can mediate the degree to which those norms take root.²¹

I argue that IOs can affect media content indirectly, through their downstream effects on actors (politicians, NGOs, and so on) with conflicting views of the underlying issue. IOs create or reignite contestation between opposing groups over the human rights framing of the issue. The media covers this contestation, shifting part of the discussion toward the countervailing messages and actions of each group. This action–reaction cycle, and subsequent media coverage, can persist even after coverage of the ICC's action itself has faded.

Consider first the effect of IOs on actors supporting the aims of the IO. Research on transnational politics describes many activities of subnational actors who want to use the language of human rights to counteract violations. They can provide information via reports or testimonies. They protest, give out awards, or invoke shame. They can lobby or leverage other governments or international institutions. Norm entrepreneurs work to provide information directly to the media.²²

IOs facilitate each of these dynamics. Simmons argues that treaties can engage previously inactive legal interest groups or “internationalist” groups to “take a new interest in the issues covered by the treaty.”²³ The same applies to supranational legal institutions like the ICC. Dancy and Montal describe how ICC actions can spur litigation from reformer coalitions.²⁴

IOs can drive attention to previously under-represented voices. A high death toll from police operations may be reported on as a crime or as a law enforcement issue. But when an organization of local lawyers is emboldened to label those deaths extrajudicial killings, their announcement may receive coverage highlighting legal and human-rights-based aspects. Each of these triggered actions can elicit media coverage, which allows these actors to advance their preferred argumentation and framing using the discourse of human rights.

Consider the following two excerpts from articles about the Philippines’ WOD.

The human rights group Karapatan hailed the ICC move, calling it a positive step towards establishing accountability and finding justice for victims of the drug war and their families. (Interaksyon, 8 February 2018)

A United Nations human rights expert urged the government to see the preliminary examination of the ICC into alleged extra judicial killings in the war on drugs as an opportunity to turn a definite page in the bloody campaign … Human Rights Watch said the ICC's announcement should spur efforts by United Nations member countries to push back against Duterte's efforts to legitimize his bloody drug war. (Interaksyon, 9 February 2018)

In both excerpts, the IO's action gave pro-accountability voices a chance to emphasize their stances in the media. In the former, Karapatan, a domestic human rights NGO, linked the WOD with human rights and accountability. In the latter, Human Rights Watch, an international NGO, emphasized similar content. The IO's action created space for a break from the usual day-to-day coverage of the issue, creating a distinct event around which human rights advocates could coalesce. These voices filled this space with their own preferred frame, casting the WOD in human rights and accountability terms.

A growing body of work theorizes how IOs trigger contestation between pro- and anti-human-rights groups, which can manifest in a variety of ways.²⁵ State actors and elites may openly resist or defy the norms espoused by an IO.²⁶ Leaders facing powerful “defiance constituencies” respond to external pressures by openly increasing their noncompliant behavior, gaining legitimacy from their intransigence. Some naming-and-shaming efforts trigger backlash from those defending “traditional” values or accusing outside and foreign actors of bias against the target country.²⁷

The media often presents both sides of a particular issue, so the amplification of pro-human-rights voices goes in tandem with the response of accused politicians and their backers. Status quo supporters and norm “antipreneurs” get to argue in favor of current policies.²⁸ The inclusion of a new actor, the IO, in the debate also opens up a front where opposing voices can criticize the IO itself. Rather than defending the status quo policy, an elite can instead shift the debate toward the appropriateness of an IO's action.²⁹ These actors engage with the human rights frame, but only to discredit it.

To again see this in practice, consider Duterte's response to the ICC. The following is from one of the articles already quoted, demonstrating how coverage of a pro-human-rights voice is often paired with coverage of the opposing view:

[Presidential spokesperson Harry] Roque also dismissed the view that the thousands of deaths in the war on drugs constituted a crime against humanity, because the war against drugs is a lawful, legitimate operation and a lawful use of force against a threat to national security. (Interaksyon, 8 February 2018)

Duterte also frequently attacked the institution itself, demeaning or threatening his detractors. The ICC, Chief Prosecutor Bensouda, and UN special rapporteur Agnes Callamard received the bulk of his ire.

Still unfazed by criticisms … [Duterte] told [Bensouda and Callamard] not to mess with him … “Go ahead, you investigate me. But I assure you—I tell them—you will never have jurisdiction over my person … If I meet them, especially the black lawyer, (Bensouda) and the slim one, Callamard, who is undernourished and not eating, do not f*** with me, girls.” (SunStar, 8 March 2018)

In related sociological research, Bail refers to this dynamic as a discursive field, which is “the public [battleground] where collective actors compete to give meaning to an issue.” The ensuing framing contests attract media attention as the actors attempt to “influence the way journalists themselves interpret or classify social issues.”³⁰ Contestation between these groups can persist well after their initial clash over an IO's action. If the IO is the spark, contestation between opposing groups is the simmering fire, and media coverage reflects its heat.

These observations imply the following hypothesis relating IO actions and the content of media coverage. An IO action can shift the discussion of an issue toward frames and topics preferred by pro-compliance actors—in this case, those related to the human rights of the victims of the WOD.³¹ This effect is amplified by the back-and-forth between actors contesting the framing of a particular issue. Even when they are no longer focused on the trigger of an IO action, human rights supporters and opponents continue their discursive contestation, which keeps human rights content in the media. Given the context examined in this study, I specify this hypothesis for an IO like the ICC, which focuses on human rights.

H1: An IO's action increases the proportion of media coverage of an issue that focuses on human rights.

Philippine Media and the ICC Examination

From a research-design perspective, the WOD and subsequent ICC actions represent a good place to assess how media coverage shifts after an IO action. Freedom House classifies the Philippines as “partly free,” with leaders chosen by open elections without accusations of widespread manipulation.³² Popular opinion can affect the government either through direct pressure or through replacement of elected representatives. Citizens can protest, and many human rights NGOs operate relatively freely. While not perfectly open, the Philippines is the type of country where an IO can feasibly mobilize meaningful action by subnational actors.

The Philippines also hosts a diverse media ecology that is appropriate for this type of study. Private media companies produce widely read newspapers, TV, and radio content. The largest broadsheet newspapers, like the Daily Inquirer and the Manila Times, also have significant online presences. Media outlets can and do criticize the government. Though outlets slant coverage along political lines, they generally cover “both sides” of an issue. The importance of the media is further demonstrated by government efforts against certain outlets. Duterte publicly campaigned against some outlets for criticizing the WOD, labeling them “fake news.” One website, Rappler, received his strongest ire, with the Philippine Securities and Exchange Commission revoking its license. Rappler continues to publish, though Duterte targeted its founder (and recent Nobel Prize recipient), Maria Ressa, with criminal charges.

In other words, the Philippine media environment is free enough to see coverage and content from competing political actors—according to the theory—yet it shares the imperfections found in many countries with an ICC situation. Overall, the Philippine media is “imperfectly free” according to the Global Media Freedom Dataset (GMFD), which is similar to other countries with ICC situations.³³ The Philippines has the same GMFD score as Kenya, Georgia, Colombia, Uganda, Libya, and the Democratic Republic of the Congo.

The Philippines has also experienced a distinct interaction with the ICC. Shortly after the WOD began, the chair of the Philippine Human Rights Commission, Chito Gascon, testified before the Senate that the WOD could fall under ICC jurisdiction as a crime against humanity.³⁴ Afterwards, the ICC was rarely discussed in the Philippines until late 2017, when the lawyer of an alleged hitman submitted a brief to the ICC. Following its practice of not commenting on specific countries, the ICC did not mention the WOD until February 2018, when it made a major announcement: the chief prosecutor had initiated a preliminary examination of the situation, a precursor to a full investigation.

In sum, the Philippine WOD constitutes an ongoing human rights issue that is the subject of a distinct action by an IO. This creates an opportunity to study changes in media coverage after the “shock” of an IO action in a place that shares many features with countries where IOs hope to have influence. Translating the hypothesis to this empirical domain: The proportion of coverage of the WOD that focuses on human rights should increase after the ICC's announcement of a preliminary examination.

Closest Related Empirical Research

A handful of studies document media coverage of IO actions in targeted states. Feinstein argues that the Special Court for Sierra Leone's outreach to the local media aided the court's mission by providing publicity for the trials and creating greater accountability for the judiciary.³⁵ Dancy finds that ICC investigations spur increased Google searches about human rights, which could direct people to media reports.³⁶ Creamer and Simmons find that in the year of a Committee Against Torture review, Latin American news outlets have approximately eight more articles that mention the committee or torture, and that this effect endures, with another three articles in the following year.³⁷ Their study is impressive for its depth and scope, covering fifteen countries over a number of years. Here, I take a fine-grained look at all the media coverage of a particular issue, seeing how the content of that coverage evolves. I zoom in on coverage of a high-salience issue, with hundreds of relevant articles per year, at a finer temporal level, to see how coverage content changes at the daily rather than yearly level.

I also focus on the possibility that increased media coverage, even when it shifts toward human rights content, might not engender public support for accountability. The content of the increased coverage matters. In the aforementioned study, Bail shows how pro-Muslim mainstream and anti-Muslim fringe groups entered the discursive media field after the rupture of 9/11.³⁸ The media's emphasis on covering “both sides” led to over-representation of the latter group's views, which they used to consolidate and deepen their influence. In the context of IOs, Brutger and Strezhnev show that media coverage of investor–state dispute settlement leans toward disputes lost by the home state, biasing readers against the settlement system.³⁹

Corpus Construction

I begin with a large corpus of 33,458 articles collected from the RSS feeds of major Philippine news outlets by the Cline Center for Advanced Social Research at the University of Illinois Urbana-Champaign. As with most electronic article databases, one challenge is limiting attention to sources and a time window where content generation is stable over time. News feeds can change which sections are included; feeds can go offline or change location. This can create gaps or changes in the data-generating process that might bias comparisons of content over time.

To examine a stable set of news feeds, I focus on four outlets—the Manila Times, the SunStar, Interaksyon, and the Philippine Daily Inquirer—and use articles published from 10 September 2017 to 1 April 2018. Within this window, these feeds produced a steady stream of articles without any obvious changes in the quantity or content of coverage. These outlets represent a large share of the Philippine broadsheet and online media market, which is dominated by English-language media. The Inquirer is the most popular broadsheet and also the most visited Philippine news website. In terms of web traffic, the Manila Times and SunStar are among the top five outlets, as well.⁴⁰ Though it would obviously be best to expand the sample time frame and the list of media outlets, this is the maximal subset with stable coverage.

Next, I used a dynamic, supervised machine learning approach to create a classifier for whether an article is relevant to the WOD.⁴¹ I first hand-coded a binary measure of relevance to the WOD for a large set of randomly chosen articles.⁴² I used each article's term frequency–inverse document frequency (TF-IDF) score to predict its WOD relevance. I then used a support vector machine algorithm to construct a classifier based on the relationship between TF-IDF data and a document's hand-coded relevance classification.

This process is most clearly explained with an example. Many articles about the WOD use the term shabu, referring to a type of methamphetamine. Among articles about the WOD, this word appears frequently: it has a high term frequency. And it is rarely found in articles that are not about the WOD: it has a high inverse document frequency. The classifier detects that an article's TF-IDF score for shabu is a good predictor of relevance to the WOD. I call an article WOD-relevant if the classifier categorizes it as relevant and it contains the term drug.⁴³ This results in very accurate out-of-sample coding.⁴⁴

Empirical Analysis of Media Content

To assess Hypothesis 1, I analyze the content of the 2,218 WOD-relevant articles using topic models, which are descriptive techniques for a corpus made up of separate documents (here, articles). Topic modeling helps infer the prominence of a theme or issue in an article, based on how words co-occur. The approach first infers a set of topics from the text. Each topic consists of a mixture of word probabilities describing the likelihood of observing that word for a particular topic. For example, the word killing might be more likely to appear in an article related to the WOD than in an article about sports.

Topic models describe each article as a weighted mixture of topics. For example, a document about Trump's comments on the Philippine WOD may be half associated with the WOD, half associated with Trump, and not at all associated with sports. For estimation, I use the well-known approach of Roberts and colleagues, setting the number of topics to seventy-five.⁴⁵

I denote the set of topics as k ∈ K and the set of articles as a ∈ A. I measure the prevalence of a particular topic on a particular day as the proportion of the coverage on day t that pertains to topic k. Let θ _a, ∈ [0, 1] denote the proportion of article a associated with topic k. Let A ^t refer to the subset of articles from day t, and let N ^t denote the number of articles on day t. I define the average prevalence of topic k on day t as

$${ \pi }_{k, t}^{av} = \sum\limits_{a{\rm \in }A^t} {\displaystyle{{{ \theta }_{a, k}} \over {N^t}}} $$

Average prevalence therefore describes the mean proportion of a particular topic across all articles on a given day.⁴⁶

I labeled two categories of topics, those pertaining to the ICC and those pertaining to human rights. Only one distinct topic covers the ICC. It emphasizes the ICC, its examination, and Duterte's withdrawal from it.

The ten human rights topics are those with human rights among their most frequently occurring terms. Articles most associated with these topics emphasize the human costs of the WOD and whether its conduct has violated human rights. Some topics are most associated with particular events, for example, when someone criticizes the WOD on human rights grounds. Others emphasize actors explicitly associated with or including human rights in their names, such as Human Rights Watch or the Philippines’ Commission on Human Rights. Topics containing the term human rights were also more likely to contain words associated with related frames, like extrajudicial killing or crimes against humanity, indicating that these are topics with an overall framing geared to human-rights-related content.

Figure 1 shows abbreviated examples of articles associated with the ICC topic (left) and one prominent human rights topic (right). The example human rights topic pertains to a series of high-profile public spats between Duterte and two officials in the UN Commission on Human Rights: Zeid al-Hussein and Agnes Caillamard. Here, too, the words with the highest probability of appearing in a topic lend credence to the coherence of the topics and my labels.

Figure 1. Example document excerpts under the ICC topic and one human rights topic

Figure 2 (left) shows the average prevalence of the ICC and human rights topics by day. Looking at the ICC topic, the ICC receives very little coverage until the announcement of the examination. And then it is a major component of coverage, but for a relatively narrow window of time. ICC topics make up over 30 percent of the WOD articles for about a week after the announcement, which is substantial. But then it fades relatively quickly. Coverage of the ICC also increases substantially several weeks after the examination announcement, when Duterte decides to withdraw from the Rome Statute. The ICC then returns as a focus, making up 40 to 60 percent of the WOD coverage and fading less quickly.⁴⁷

Figure 2. Effect of ICC on human rights and ICC topics, September 2017 to April 2018

More importantly, how did the prevalence of human rights topics change after the ICC's examination announcement? These figures show systematic evidence that actions by IOs can indirectly affect the media conversation about violations of international criminal law. Overall, coverage of the WOD pertaining to human rights topics fades over time. Human rights topics decline around Christmas 2017, which also corresponds to a general decrease in coverage of the WOD during the holidays. Human rights coverage declines steadily and reaches its lowest point in the month before the ICC announcement.

However, after the ICC examination announcement, human rights returns to greater prevalence in media coverage—distinct from any coverage of the ICC itself. In the next weeks, the proportion of WOD coverage pertaining to human rights topics reaches its highest level over the whole time window. On average, 15 to 25 percent of the WOD coverage after the ICC announcement pertains to human rights, a level not sustained since the earliest parts of the analysis window.

Which aspects of human rights coverage increased after the ICC's announcement? Of the ten human rights topics, only four increased in prevalence; and these four emphasize contestation between human rights advocates and Duterte (Figure 2, right).

The topics that spike after the ICC examination (UN Spats and UN CHR/HR Report) pertain to a highly public spat between Duterte and UN officials, notably the high commissioner for human rights, Zeid Raad al-Hussein.⁴⁸ On 1 March 2018, Duterte instructed police to ignore the UN's special rapporteur, Agnes Callamard. Later that week, his spokesman invited her to “swim in the Pasig River.”⁴⁹ On 7 March, al-Hussein condemned Duterte in a speech before the Human Rights Council.⁵⁰ During a later news conference, he suggested that Duterte needed “psychiatric evaluation.” Duterte responded by calling al-Hussein an “empty-headed … son of a wh***.”⁵¹ The HRW/Roque topic has to do with a public back-and-forth between Spokesman Roque and Human Rights Watch representatives. The Mixed topic is about the WOD in Cebu; after the ICC announcement, this topic captures the Cebu City mayor's criticism of the Human Rights Commission.⁵²

While the trends in the topics are apparent on visual inspection, Figure 3 shows a statistical assessment of topic prevalence before versus after the ICC's examination announcement.⁵³ Each point (and corresponding 95 percent confidence interval) shows the expected change in topic prevalence for an article after the ICC's announcement compared to before. Appropriately, there is a large and statistically significant increase in ICC coverage. The increase in coverage of spats between Duterte and human rights officials is also large and significant. The HRW/Roque and Mixed Human Rights topics also show increases, although the estimated effect of the ICC announcement is not as strong.⁵⁴

Figure 3. Topic prevalence before/after ICC, structural topic model estimates

Inference Challenges and Robustness Checks

Exogeneity

I have treated the ICC examination as an exogenous shock to media coverage. It is possible that coverage of the WOD anticipated the ICC's examination, or vice versa. In the former case, media members might have adjusted their coverage in anticipation of an ICC announcement. In the latter case, the ICC might be strategically timing its announcement based on changes in public opinion on the WOD or changes in the favorability of media coverage.

However, both scenarios are very unlikely. In the appendix, I describe qualitative and quantitative evidence suggesting that reporters knew of an ICC announcement only a few days in advance and held their stories until after the announcement. Public opinion data also show that Duterte and the WOD remained popular before and after the ICC's announcement.⁵⁵

Placebo Testing

It is possible that spikes in the human rights content of WOD coverage simply occur over time, and they might coincidentally occur after the ICC's examination. In the appendix, I consider in greater detail three pieces of evidence that support my interpretation of the results, that an IO triggered contestation which shifted coverage toward human rights topics: (1) I can trace the relationship between post-ICC topic spikes and the ICC; (2) other spikes in human rights coverage are associated with IOs; and (3) spikes like those focused on here are not common. I also show that another major actor, the Catholic Church, does not trigger coverage of contestation in the same way as the ICC. I also show that coverage of human rights outside of the WOD context does not increase after the ICC's actions.⁵⁶

Number of Topics

I chose the number of topics ex ante, with a desire to have fine-grained topics that were still recognizable. There is not a statistical quantity that identifies the “correct” number of topics. In Appendix G, I used an algorithm that first selects the number of anchor words for topics, and therefore the number of topics, from a lower-dimensional space, which can improve topic coherence and interpretability. I found trends that were similar to those described earlier.

Survey Experimental Evidence

Does shifting media coverage affect attitudes to the WOD? Here, I briefly present survey experimental evidence that coverage of the ICC—without content emphasizing contestation—lowers support for the WOD. This is, initially, “good news” about the effect of the IO, from the perspective of the ICC. However, coverage of contestation over the ICC blunts this effect, restoring support for the WOD to its initial levels.

Survey Design

I used a professional firm to recruit approximately 3,000 respondents from the Philippines, and the survey was fielded 21 September to 21 October 2021. Respondents were recruited so that the sample was nationally representative along several demographic characteristics. They participated online and took the survey in Tagalog. Native speakers checked the translations for all materials. The research was preregistered at Evidence in Governance and Politics (EGAP).⁵⁷

I constructed three media articles that captured the key features of shifting coverage documented earlier. Each treatment article consisted of excerpts from real articles from prominent Philippine outlets, with light editing to combine them.⁵⁸ I constructed each article to match the archetypes highlighted in the theory and media content analysis: a “regular” article about the general conduct of the WOD (WOD Article), an article about the ICC investigation that emphasized contestation (ICC Contestation Article), and an ICC article that emphasized facts about the investigation in lieu of contestation (ICC No-Contestation Article).

The WOD Article focused on the everyday operations of the drug war and the overall statistics on its impact. This type of article has been common across the time frame of the analysis and since. These articles tend to describe a specific drug operation or incident, often ending with summary information about the total number of people killed. They do not generally contain judgment about whether the WOD has been good or bad.

The ICC No-Contestation Article resembled articles reporting directly on the ICC's actions. In the time frame of the survey, the ICC had recently announced its transition from a preliminary examination to a full investigation. These articles describe what the ICC did, its process, and the basis for the ICC's actions.

The ICC Contestation Article resembled a common type of article which reported the ICC's actions and then emphasized contestation—the back-and-forth exchanges between supporters and opponents of the WOD. These were between human rights groups, opposition politicians, and representatives of IOs on the one side, and Duterte and his supporters on the other.

The three treatment articles had similar length and style, so the main difference was in the content. Respondents were randomly assigned to read one of these articles or to a control condition where they did not read an article.⁵⁹

After reading an article or being assigned to a control condition without an article, respondents answered a question: “Do you support or not support the campaign against illegal drugs?” Available responses were truly support; support; may support and may not support; not support; and truly not support. This wording matches that used in the public opinion surveys conducted by major Philippine polling firms.⁶⁰

Analysis

For analysis, I used a binary indicator of support for the WOD, coded 1 for “truly support” or “support” and 0 otherwise. I regressed this (OLS) on indicators of treatment assignment. I also included an indicator variable coded 1 for respondents who said they were politically closest to the Liberal party and 0 otherwise. I interacted this with the treatment indicators, to let treatment effects vary by party identification.⁶¹

Figure 4 shows the predicted probability of supporting the WOD by party and treatment condition. Moving from a “regular” WOD article to an ICC article without contestation lowers support for the WOD, primarily because of the substantial drop in support among Liberal Party supporters. Their support decreases by approximately 12.5 percent. It also leads to a slight drop in support among supporters of other parties. In this sense, coverage of the ICC has its desired effect, from the perspective of the court.

Figure 4. Predicted support for the war on drugs, by treatment and party support

However, contestation also “works,” from the perspective of the WOD's proponents. Shifting from an ICC article without contestation to a similar article with contestation re-raises support for the WOD by approximately 8 percent among Liberal Party supporters. There is a slight additional decrease among supporters of other parties, but it is more than offset by the restoration of support among Liberal Party supporters. The net result is that support levels are similar to those observed among respondents who read a “regular” WOD article.

It is possible that contestation simply weakens Liberal Party supporters’ opposition to the WOD, though it is unlikely that Duterte's opponents are persuaded by his rhetoric. It is perhaps more likely that coverage of contestation displaces other information about the ICC's actions that would have depressed their support even further. The absence of additional, more neutral content about the ICC potentially explains why the ICC Contestation article does not decrease support for the WOD nearly as much as the ICC No-Contestation article without contestation.

Conclusion

Many core arguments describe how IOs’ actions change public opinion and mass politics. This study provides systematic analysis of how IOs affect national media, which is a crucial interlocutor between IOs and the public. I find that the proportion of media coverage of the WOD emphasizing human rights content increases after the ICC's action, mainly from increased coverage of contestation between groups supporting and opposing the WOD.

The results help us better understand the potentially subtle, positive effects of IOs and also the tough sledding they face in shifting public opinion. The last few years have been discouraging for IOs, especially the ICC. Facing populist, right-wing, and anticolonial blowback, many institutions have struggled to deepen or widen compliance. Their inability to convince large subsets of the general population of the worthiness of their goals has limited their ability to put political pressure on actors committed to noncompliance. Media coverage of IO actions gives one potential explanation. An IO action might help amplify the voices of pro-compliance efforts, but this is framed largely in dialogue with vocal resistance from opposing actors. And as the survey experimental evidence here suggests, positive coverage of the court decreases support for problematic policies like the WOD, but replacing this with two-sided coverage mutes those positive effects. If coverage of IO actions gives prominent voice to IO detractors, then this could counteract the bottom-up effects of actors supporting human rights.

In the shorter run, IOs may be unable to avoid this. IOs like the ICC face an “inescapable dyad” with respect to political engagement.⁶² If an IO “stays out” of public opinion battles to retain the appearance of impartiality, then it cedes a critical arena to potential opponents. But if it actively engages in this arena, then it will be charged with “playing politics” and abandoning its objectivity. The ICC has generally chosen the first approach. It makes some local outreach efforts, but these are inconsistent and limited in scope, though the court has recently announced that it will hire multiple “country experts” to liase with local media. If the ICC and other IOs change tack and lean into efforts to persuade the public, then ongoing and future research that will help us understand the conditions under which different arguments, messages, and frames influence opinions will be critical,⁶³ as well as research on how governments choose to parry these efforts.⁶⁴

Research on the effects of coverage on public opinion in the longer run is also important. Initial contestation and its coverage in the media can plant seeds that later bear fruit. A pro-human-rights voice may react to an IO's action only to be drowned out in the pages of the newspaper by opposing groups, yet that voice may still gain traction gradually. Alternatively, initial coverage frames may prove sticky, and IO actions could harden opinions across different segments of the public. This study has provided an approach to looking for those longer-term effects. Doing so would require a longer temporal coverage of media content and public opinion from before and after key events.

Future research could also examine the conditions under which IO actions result in coverage emphasizing contestation, as opposed to more one-sided or pro-IO coverage. This study focused on one action (a preliminary examination) by one IO (the ICC) in one country (the Philippines). This is a good entry point for this type of research, since it is emblematic of many settings where IOs seek to have an impact: democratic countries with partially free media. Yet media coverage could differ as any of those features—the action, institution, or setting—change. For example, coverage of post-electoral violence in Kenya emphasizes the ICC at an early stage, and ICC actions neither spur nor supplant additional coverage of human rights.⁶⁵ Coverage of regional institutions, like the Organization for Security and Co-operation in Europe or the Inter-American Commission on Human Rights, might be less emphatic of voices bemoaning foreign interference. In a more closed political setting, like Sudan, IO actions may fail to move media coverage at all. Or they may be the necessary spark that opens the door for previously closed venues to host pro-human-rights voices. In more open polities, IOs may only magnify voices that already occupied a large part of public discourse. As work on IO politicization has highlighted,⁶⁶ even lower-salience IOs like the World Trade Organization are being dragged into public spats. It is not hard to imagine a future discursive field where a WTO ruling against a steel tariff, for example, could amplify the messages of free trade firms and politicians espousing frames related to the economic gains from globalization or the importance of legal rules. Their messages would likely be paired with those of protectionist actors lamenting the loss of native manufacturing or economic dislocation.