Knowing your audience: How the structure of international relations and organizational choices affect amnesty international’s advocacy

Abstract

While research has addressed the effects of international nongovernmental organizations (INGOs) advocacy on human rights outcomes, less is known about how INGOs choose advocacy targets and tactics. We combine insights from political economy and constructivism to understand how INGOs come to choose targets and tactics through the concepts of information and leverage politics, first articulated by Keck and Sikkink (1998), and salience politics, or the need to select cases that energize organization members and donors. INGOs select potential targets for advocacy and choose their tactics based on considerations of leverage potential and political salience, both of which are a function of potential target states’ aid, trade, and security linkages with major Western powers. Using data on Amnesty International’s written advocacy efforts - background reports, press releases, and new data on Urgent Actions - we find robust evidence that Amnesty International accounts for these linkages with Western powers in choosing targets for its advocacy campaigns.

What factors contribute to “naming and shaming” choices made by international nongovernmental organizations (INGOs)? Human rights INGOs commit themselves to revealing the violations of the world’s human rights offenders as “witnesses” and exposers of human rights abuses (Hopgood 2006, Redfield 2006). Yet, as strategic actors, they make difficult choices regarding the issues and cases to which they can devote limited resources. The attempts of Amnesty International (AI), perhaps the foremost human rights advocacy organization, to name and shame have received the most scrutiny (Clark 2001, Poe et al. 2001, Ron et al. 2005, Hafner-Burton 2008, Hill et al. 2013). AI is a central hub of the international human rights movement and has a long history of reporting on human rights issues (Ron et al. 2005, Lake and Wong 2009, Murdie and Davis 2011, Wong 2012). Most extant work, however, focuses on the effects of naming and shaming in affecting human rights conditions in targeted countries. We know comparatively little about why AI chooses to focus on particular countries and cases at the expense of others, and what factors affect the form this advocacy takes.

In 1997, civil war raged between former President Denis Sassou-Nguessou’s Cobra militia and government forces loyal to then President Pascal Lissouba in Brazzaville, Republic of Congo. The fighting saw interventions by the Angolan and Chadian armed forces and included artillery shelling of the capital city, leading to as many as 20,000 battle deaths in just that year.¹ These high levels of political violence and state repression were reflected in the assessments provided by both the Physical Integrity Rights and Political Terror Scales, receiving the worst scores possible.² However, AI issued no background reports or press releases – its primary forms of human rights reporting – and only called for a single letter writing campaign (Urgent Action, or UA). In contrast, Mexico was covered by seven background reports, nine press releases, and 58 UAs that same year, mostly in response to repression of civilians thought to be supporters of the Ejercito Zapatista de Liberación Nacional (EZLN), also known as the Zapatistas. While this repression included acts of horrific violence, such as the massacre of 45 pacifist indigenous rights activists by a government-aligned paramilitary group at Acteal, Chiapas, the smaller scale and breadth of abuses in Mexico, and the population sizes of the countries (Mexico is 27 times larger than Republic of Congo), do not entirely explain the discrepancy in targeting.

We unpack the concept of naming and shaming, demonstrating that INGOs have choices whether to, and how, they speak out against human rights abuses. In doing so, we focus on organizational choices at the campaign level to demonstrate how human rights INGOs use different advocacy tactics to pursue competing goals. Human rights INGOs, and AI in particular, attempt to simultaneously 1) serve as “impartial” observers and chroniclers of human rights abuses, 2) try to rally powerful states to action against violators, and 3) serve the interests of their stakeholders and engage the passions of their members. INGOs thus address different audiences, hoping to motivate responses from a variety of political actors. We follow earlier work to demonstrate how INGOs might balance principled (Sikkink 1993, Keck and Sikkink 1998) and materialistic (Cooley and Ron 2002, Werker and Ahmed 2008) motivations through an evaluation of AI. Understanding the complexities of its decision-making provides a blueprint for studying other INGOs in human rights and beyond, given its organizational prominence and centrality to global human rights advocacy. Scholarship on naming and shaming has progressed to the point that we need to understand not just its effects, but how INGOs go about choosing targets and tactics.

We develop a theory of how INGOs choose advocacy tactics in part based on aid, trade, and security linkages between violator states and powerful Western states. Building on earlier work on information politics and leverage politics (Keck and Sikkink 1998), we advance the idea of salience politics: INGOs target countries and issues for coverage because they are important to the organization’s membership and funders. Salience politics highlights the value INGOs place on mobilizing individuals and donors in order to satisfy their resource needs. INGOs, like other political organizations, must survive if they wish to affect policy outcomes. Leverage politics, by contrast, focuses on INGO efforts to influence powerful Western states to mobilize around the INGO’s cause. Information politics is built around developing a reputation as a credible, nonpartisan source of reporting on human rights abuses.

Aid and security relationships with powerful Western states provide both leverage and salience, as they constitute highly visible policy choices that confer implicit legitimacy and resources on foreign governments, and Western governments can impose conditions on aid and security assistance in order to elicit behavioral change from violator states. Trade relationships, on the other hand, confer salience but not leverage. Trade relationships may generate heightened awareness of other countries, but for reasons both ideological and political-economic, they do not necessarily provide a source of leverage to be manipulated by Western states. Depending on the goals of a particular campaign, INGOs choose to employ tactics that reflect 1) information politics, in which case, reporting should not be disproportionately aimed at countries with linkages to powerful Western states, 2) leverage politics, in which case, targeting should be correlated with aid and security relations with powerful Western states, and 3) salience politics, in which case, targeting should be correlated with aid, trade and security relationships. Often, INGOs will engage in a mixture of all three. Our argument, therefore, explores how INGOs choose different tactics in pursuit of different – and potentially competing – organizational goals.

The AI tactics we explore here are background reports, press releases, and UAs. Background reports document the extent of abuses in meticulous detail, whereas press releases are designed for quick dissemination and consumption. UAs require AI member involvement, so the Secretariat faces strong incentives to issue them for countries that are salient to the membership in order to solicit participation. This allows us to make inferences about the types of international linkages that should be important to AI rank-and-file members, and to discern whether different types of linkages matter for different types of AI targeting: whether background reports, press releases, and UAs tend to follow similar or different patterns of linkages. In developing our hypotheses, we focus on the primary audiences for each form of advocacy, acknowledging, for instance, that even though background reports are intended mostly for policymakers and issue specialists, anyone access these documents.

Controlling for the level of abuses and population, two variables that most clearly proxy objective need for advocacy, we find aid, trade and security linkages with the USA, and to a much lesser extent the UK, affect targeting for the various forms of advocacy in ways that are both consistent with and challenge our theoretical expectations. While aid, trade and security linkages have only weak or insignificant effects on targeting for background reporting, both trade and security linkages exert effects on targeting for UAs. Interestingly, only security linkages matter for targeting for press releases. Moreover, robustness checks indicate these findings are not an artifact of the USA having closer relations with Latin America, a region that has received considerable attention from AI (Clark 2001, Hafner-Burton and Ron 2012). The quantitative evidence suggests that AI uses multiple forms of advocacy because doing so helps the organization balance competing internal interests: maintaining its reputation as an unbiased source of information, attempting to curb human rights abuses, and keeping its membership engaged.

This article makes two contributions. First, we offer a theoretical way to reflect on different INGO advocacy tactics by introducing the idea of salience politics as a complement to information and leverage politics, and pointing out that different advocacy tactics are aimed at different primary audiences. We consider a broad range of linkages that includes aid, trade, and security relationships, such as military transfers and alliance membership. Second, the analysis of new data on UAs, which require active member participation, allows us to investigate the preferences of the rank-and-file members that make up AI’s primary source of funding. UAs are an important way for AI to engage local NGOs as well (Meernik et al. 2012). While not all INGOs have large, active memberships, all INGOs relying on external sources of funding face conceptually similar constraints on their autonomy.

Even though the data are focused on AI, we expect that the logics of target and tactical choice hold for any actor facing potential tradeoffs between serving as a credible source of information, engaging in leverage politics, and motivating their membership and funders to participate. We conclude by discussing the broader relevance of our findings for the study of non-state actors in international relations. Given that INGOs are finding increased legitimacy and credibility as providers of information for policymaking (Clark et al. 1998, Gourevitch et al. 2012), these findings offer a way to theorize systematically about INGO reporting bias.

1 Explaining INGO advocacy

Comparatively few scholars have looked directly at INGOs from an organizational perspective, which is a view that emphasizes the structures of internal decision-making, and why organizations do what they do (Cooley and Ron 2002, Werker and Ahmed 2008, Stroup 2012, Wong 2012; Murdie forthcoming). Keck and Sikkink (1998) emphasize that transnational advocacy networks, in which INGOs are major actors, are principled actors with instrumental goals that they pursue through information politics, symbolic politics, accountability politics, and leverage politics in the boomerang pattern.³ Understanding the internal motivations that lead to tactical choices will help us better grasp naming and shaming as an instrumental and principled process. INGOs do not approach every human rights violation, or every country, in the same way. We contribute to a growing discussion that stems from both the constructivist and political economy literatures.

Our approach draws on the internal political economy of INGOs themselves. INGO behavior is conditioned by incentives to cultivate credibility and demonstrate efficacy in order to generate operating capital. Cooley and Ron’s (2002) seminal piece on non-state actors demonstrates that INGOs respond not just to their moral imperatives, but to economic incentives, in some cases neglecting their “official” reasons for being in the field in order to establish a public image or seek continued funding. Others have extended analogies of the firm or interest groups to highlight the material interests of INGOs (Johnson and Prakash 2007, Prakash and Gugerty 2010, Bloodgood 2010). For instance, case studies demonstrate that AI and Human Rights Watch neglected economic, social, and cultural rights in favor of civil and political rights for organizational reasons up until the past decade (Roth 2004a, Chandhoke 2007, Goering 2007). Defending these choices, INGO leaders often cite limited resources and the need to be able to build a causal story around violations in order to generate internal support. As competition for funding and influence have increased, scholars have increasingly modeled INGOs as strategic actors that face resource-based incentives and constraints on their actions (Reimann 2006, Werker and Ahmed 2008).

2 Theory: A model of how human rights INGOs think tactically

2.1 Why different tactics matter

Because INGOs have limited resources, they face tradeoffs in reporting on different countries and cases of human rights abuse. When selecting a case for advocacy, INGOs must weigh the severity of violation(s) against beliefs about how an advocacy campaign is likely to 1) change violator behavior and 2) affect the organization itself. These two factors will affect many if not all INGO decisions, as INGOs both want to effect political change and survive as organizations.

Our analysis starts with two insights. First, INGOs are principled actors. Second, INGOs face material incentives and constraints. Similar to Cooley and Ron (2002), we conceptualize INGOs as organizations with specific objectives that may be at odds with recipients and supporters, and like Ron et al. (2005) and Hill et al. (2013), we are interested in explicating what shapes the advocacy choices INGOs make. These contributions emphasize the need for acknowledging the material constraints facing morally driven actors, such as human rights INGOs.

Cooley and Ron argue the proliferation of INGOs and donors in international politics has led to a conundrum: INGOs must serve both the populations they have identified as needing help (recipients) and their donors. INGOs often receive a mix of funding from foundations, states, and intergovernmental organizations (IGOs). INGOs are in part beholden to these (often) large donors, many of whom have set increasingly stringent limits on what INGOs can do with their funds (Barnett 2005, 2009). In some cases, as with AI, there is the additional constraint posed by membership. Because AI has generally refused donations from governments and foundations, it has relied upon its membership for financial support (Welch 2001, Winston 2001). The organization’s chief executive body, the International Secretariat in London, is funded through the contributions of its 56 national sections, with most operating capital coming from Europe and North America.⁴ The members also form part of the governance structure and participate in local groups and writing UAs.

One comprehensive quantitative treatment of AI targeting is Ron et al. (2005). Controlling for the extent of human rights abuses (as reported by AI), population, and other factors, they found some weak evidence that countries receiving more US military aid get targeted more frequently for background reports, but not press releases. Countries that receive more official development assistance (ODA) are somewhat less frequently targeted for background reports. In general, economic and security linkages with the USA are not significant determinants of targeting for press releases. Hill et al. (2013) show that while AI may have incentives to inflate its claims of abuses, AI tends to overstate the extent of abuses only when there is already substantial media attention. This finding is consistent with our intuition that AI, similar to other INGOs, acts strategically in response to what other actors (e.g., states, IGOs, media, other INGOs) do. This article understands INGOs as interacting with and shaping their environments through their actions (Aldrich 1979).

Research to date points to the role that audience plays in shaping the tactical choices of INGOs; the interaction between audience and INGO shapes a large part of whether and why an INGO is seen as credible (Gourevitch and Lake 2012). The audience for leverage politics is not the violator state itself, but other actors (i.e., states, the media, international organizations). Leverage politics operates through the boomerang pattern (Keck and Sikkink 1998), in which INGO advocacy substitutes for ineffective advocacy at the domestic level in a violator state through the lobbying of IGOs and/or powerful states in Europe and North America. This implies that INGOs are strategic in their selection of targeting for advocacy, and should focus attention on violator states over which Western governments have influence. In contrast, the audience for salience politics is the organization’s members and funders. In choosing targets for advocacy, INGOs – and this includes the media – have incentives to target those cases that are already highly salient to their audience.⁵ In this context, “salient” means that members and funders are familiar with a country/case and will be more motivated to act around it, thereby continuing to provide the organization with resources (see Mani and Mukand 2007, Cao and Prakash 2012).

We are concerned with whether countries are salient to rank-and-file members – and to a lesser extent, ordinary citizens – who would potentially mobilize in support of AI’s work. While we acknowledge an overlap between leverage and salience politics and that one may affect the other, we also think it critical to distinguish between the two political mechanisms because of the different primary audiences that they target. There is no reason to think that background reports, press releases, and UAs cannot work together – in fact, they often are employed simultaneously – but they operate according to different political logics, and these different logics illuminate the sometimes difficult choices INGOs face.

2.2 Thinking like an INGO

If INGOs are both material and principled actors that work strategically to advance their political goals, do their choices of tactics and targets reflect these priorities? INGOs may wish to effect immediate changes in violator-state policy, but also have incentives to build a reputation for providing credible information (see Gourevitch et al. 2012). If they become reliable sources, INGOs can enhance their status. However, the imperative to provide unbiased reporting is at odds with the notion of leverage politics, which implies attention should be focused on countries with ties to powerful Western states. Disproportionately targeting states because of their ties to Western states, however, carries the risk that the organization will come to be viewed as partisan and biased, undermining its credibility. If an organization can create a distinction between its efforts as a reporting agency (background reports) and an advocacy organization (press releases and UAs), it may be able to pursue these competing goals simultaneously.

Furthermore, it is important to keep organizational limitations in mind. INGOs are limited in the scope of their work by funding opportunities (Cooley and Ron 2002, Werker and Ahmed 2008), organizational structure (Wong 2012) and constraints from domestic contexts (Stroup 2012). How decisions are made within an INGO, and the stakeholders involved in the agenda-setting process, help determine not only what an INGO advocates (Shawki 2011, Bob 2012), but what tactics they adopt for that advocacy.

In implementing information, leverage and salience politics, three types of international linkages will matter for INGOs: aid, trade, and security linkages with the USA and the UK. We focus on links to the USA and UK for two reasons. First, the concept of leverage politics emphasizes the role of powerful Western states (Keck and Sikkink 1998). The USA has been the most powerful actor in the international system for the entirety of AI’s existence, while the UK’s vast former empire has left it with significant economic and security ties to many parts of the developing world. Both the USA and UK have active human rights policies, with London viewed as a principal hub for human rights work, especially by British INGOs, such as AI (Sikkink 2004, Stroup 2012: 161, Stroup and Murdie 2012: 438–440). AI was “born” in London, and the AI-UK office has historically played an influential role (see Hopgood 2006). Second, the USA and the UK are home to two of the largest membership sections of AI in terms of funding (Stroup 2012: 149) and membership, and as such, their aid, trade, and security relationships with other countries are likely to be highly salient. Given the preeminent position of the USA in the international system, we expect US linkages to be a more significant determinant of AI behavior.

Aid, trade, and security linkages should affect the frequency of targeting in ways that vary across the methods of advocacy. Linkages can affect both country leverage and salience potential. Salience is a function of familiarity with and concern for a case, and should matter for advocacy efforts that try to engage the organization’s membership. Leverage potential is largely a function of a country’s dependence on the USA or UK for security and/or economic development, and should matter primarily for advocacy that attempts to employ leverage politics.

Figure 1 provides a guide for conceptualizing the different types of linkages and their salience and leverage potential. All four types of international linkages should increase salience: aid (as ODA), trade, and security linkages (alliance membership and arms transfers). AI members in the USA and UK are likely to be more sensitive to human rights abuses occurring in countries that are allies of these countries. Alliance patterns reflect diplomatic and security relationships that explicitly confer legitimacy on foreign states. Arms transfers and ODA represent highly visible policy choices that receive significant reporting in the USA and UK, and which motivated AI members in the past. AI’s focus during the 1980s on small Latin American countries like El Salvador and Guatemala was driven in large part by those countries’ dependence on US military and development aid (Carleton and Stohl 1985, Clark 2001, Hopgood 2006). Regarding trade, international trading partners are the subjects of more news reporting in general (Wu 2000), and country-of-origin labeling makes every trip to the supermarket or big box store a reminder of a country’s economic ties to foreign countries, especially for individuals who have self-selected into membership in a human rights INGO.⁶

Fig. 1

A conceptual framework for international linkages, leverage, and salience

In terms of leverage potential, it is relatively straightforward to assert that the USA and UK should, ceteris paribus, have more influence over their allies than over countries with which they have no alliances and whose positions in multilateral organizations, such as the UN, are generally in opposition to those of the USA and the UK. Similarly, both ODA and arms transfers require explicit government approval (arms transfers) and/or action (ODA). Western countries have used negative aid conditionality – the suspension of ODA transfers in response to human rights abuses – and positive aid conditionality – providing more ODA after improvements in human rights performance – to curb abuses (Poe 1992, Rich 2004, Carey 2007). Examples of explicit issue linkage between arms sales and human rights conditions include the European Union’s embargo of arms sales to China following the Tiananmen Square massacres of 1989, the US Leahy-Feingold Amendment of 1994,⁷ and the US Child Soldier Prevention Act (CSPA) of 2009, which ostensibly prohibits arms transfers to those countries whose militaries employ child soldiers.⁸ Furthermore, arms transfers and ODA have direct effects on the ability of governments to continue abuses. Arms transfers provide the weapons violator states need to repress their populations. ODA generally confers significant political benefits on the state actors that control its targeting and expenditure. For these reasons, explicitly linking human rights to arms transfers or ODA provides states with the incentive to comply with human rights norms. Thus, alliance patterns, arms transfers, and ODA should increase leverage potential.

By contrast, trade linkages should increase country salience but should not confer significant leverage potential for two reasons. First, the suspension of trade relationships, much more so than ODA or arms transfers, entails large economic and political costs for both the target country and the country imposing the sanction. These costs increase with volume of bilateral trade. Second, economic sanctions typically have drastic and negative effects for civilian populations in targeted countries (Ascherio et al. 1992, Garfield et al. 1995, Lopez and Cortright 1997, Ali and Shah 2000, Cortright and Lopez 2000), while having weak or indeterminate effects on targeted countries’ policy resolve, even resulting in occasional authoritarian backsliding (Tsebelis 1990, Pape 1997, 1998, Lacy and Niou 2004, Marinov 2005, Peksen and Drury 2010). In particular, economic sanctions may worsen human rights conditions in targeted countries (Peksen 2009). These effects have not gone unnoticed among INGOs: historically, AI has not been in favor of trade sanctions because of worries that such policies mostly harm the very populations they seek to protect.⁹

To summarize, the concepts of leverage potential and country salience can be linked to specific forms of AI advocacy. We hypothesize about the importance of leverage and/or salience politics with respect to different types of advocacy by identifying the primary audience for a given tactic. This closer look at the different tactics AI uses not only offers more precision with our theories of how organizations behave, but also reveal how INGOs think strategically, and as such, moves the literature beyond Ron et al.’s (2005) analysis of the macro-level factors that shape the decision to advocate.

AI uses three main advocacy tactics: 1) background reports that document and “witness” abuses, 2) press releases to build awareness through the media, and 3) grassroots mobilization through letter-writing (UAs). The incentives to target countries with higher perceived salience to AI members and funders and greater leverage potential for the USA and UK should differ across the various forms of advocacy, according to the primary organizational goal that each type of advocacy is intended to further.

According to our theory, background reporting represents AI’s attempt to engage in information politics: providing detailed information about abuses to a core audience of policymakers and other specialists. Because of AI’s central role in the human rights network, many INGOs, government agencies, and scholars rely on AI’s reporting of abuses. The value of this reporting is related to its perceived impartiality; in order to preserve its status as an impartial chronicler of human rights conditions, AI should not disproportionately target countries with linkages to major Western powers.

• H1: Neither aid, trade, nor security linkages with the USA and the UK will be associated with the frequency with which a country is targeted for AI background reporting.

Press releases are emblematic of AI’s use of leverage politics because they primarily attempt to sway press attention toward particular governments and abuses. They are short, highlight key issues, identify wrongdoers and victims, and explain what needs to be done. According to the logic of the boomerang pattern, press releases represent attempts to engage in leverage politics, activating media and elite interest in Western states. Thus, targeting for press releases should correlate with linkages that confer leverage potential: alliance patterns, arms transfers, and ODA. As press releases are issued directly by AI’s headquarters and do not require membership participation, they should be affected only by those linkages that confer leverage potential. They should not be correlated with bilateral trade, which confers only salience.

• H2: Aid and security linkages with the USA and UK will be positively associated with the frequency with which a country is targeted for AI press releases.

UAs are distinct from background reporting and press releases because they require the active participation of AI members in letter-writing campaigns; the primary audience of UAs are the rank-and file AI members, and to a lesser extent, ordinary citizens.¹⁰ The raison d'être of UAs, in fact, was to find a way to economically engage a broad public in the politics of human rights. UAs are written in such a way to be accessible to non-specialists and non-decision makers who want to get involved. Thus, following the logic of salience politics, AI has incentives to target countries with higher salience more frequently in order to engage its members and the public. Thus, UAs should correlate with all those linkages that confer salience.

• H3: Aid, trade, and security linkages with the USA and UK will be positively associated with the frequency with which a country is targeted for AI UAs.

3 Empirics

3.1 Dependent variables

We estimate the effects of aid, trade, and security linkages with the USA and UK on three different types of AI advocacy: background reports, press releases, and UAs. In decreasing order, background reports, press releases, and UAs reflect varying levels of primary research. Background reports are lengthy, research-based documents that are written primarily for a specialized audience of government officials, INGO officials, and academics. Within AI, these documents are held to high evidentiary standards, and there can be a significant time lag between abuses and publication (Stroup 2012). Reports often contain extensive interviews, attempts to generalize about the situation, and policy recommendations. We theorize that they are the main way AI engages in information politics. Press releases, meanwhile, are intended to cover and shape responses to current events, and necessarily are based around less rigorous reporting. The primary audiences for press releases are Western governments, the international media, and IGOs. We theorize that they are the main way AI engages in leverage politics. Data on background reports and press releases were coded from the Amnesty International Cumulative Guide 1962–2000; while the original coding covered the period 1986–2000, the data have been coded back to 1975 (Ron et al. 2005, Hafner-Burton 2008).

We use also a new dataset of 12,865 UAs targeting 171 countries for the period 1975–2004 (Hendrix and Wong 2013). Letter-writing campaigns aimed at violator governments have long been AI’s claim to fame. The Secretariat issues UAs to national sections, based on assessments by the relevant regional research teams. Each UA contains a brief description documenting the name of the individual whose rights might have been abused, all known circumstances surrounding his/her abuse, the government or actor that perpetrated the violation, and detailed contact information for relevant violator state officials. Multiple UAs may be issued for a single individual; multiple individuals may be covered under any given UA. These dossiers are distributed at the national level to interested members in the UA Network, who then write letters in immediate response to appeal the violations detailed in the dossier.¹¹ We theorize that UAs embody AI’s most direct attempt to engage in salience politics.

The country-year mean of UAs is 2.42, though the data are skewed, with 40 % of country-years not being targeted at all, and only 5 % of country-years being targeted more than ten times. AI background reports, press releases, and UAs are relatively strongly correlated. This is to be expected as all three are measures of the concerns and priorities of the same organization. However, not all countries targeted for UAs are also targeted for backgrounds reports and/or press releases and vice versa. Figure 2 shows the top 10 targets for AI advocacy during 1975–2000. While most of these countries are populous and have poor human rights records in general, this is not uniformly the case. Guatemala, a country of fewer than 15 million people, received more attention from AI between 1975 and 2000 than China. The USA is in the top quartile worldwide for compliance with international human rights standards over this period,¹² yet is the second most-often targeted country. This is due almost entirely to the USA’s use of capital punishment (Thompson 2008).

Fig. 2

Top ten targets of Amnesty International background reporting, press releases, and UAs, 1975–2004

To illustrate how the composition of AI advocacy can vary across cases, during the First Intifada, AI made Israel one of its highest priorities for human rights reporting. In 1991 alone, AI issued 67 background reports on human rights conditions and violations in Israel and the Occupied Territories – second only to the collapsing and chaotic Soviet Union. Clearly, Israel was a high priority for AI. Yet, that same year, AI issued only two UAs and one press release on Israel. At the other end of the spectrum, AI issued no background reports as the Democratic Republic of Congo descended into widespread violence in 1996–1997, but singled out that country for 32 UAs and 22 press releases.

3.2 Independent variables

Our key independent variables are measures of aid, trade, and security relationships. We operationalize aid flows as the log-transformed annual dollar amount of ODA from the USA and UK to potential target countries (Findley et al. 2009). ODA consists of uncompensated transfers from donor governments and aid agencies to developing country governments that are intended to promote economic development and welfare. It does not include military aid. We operationalize trade using trade flows. We do not include measures of participation in free trade agreements, as bilateral and multilateral agreements that address trade do so ostensibly to increase flows between countries. To the extent that these treaties have effects that do not operate through trade, these effects will not be captured. In our preliminary analyses, we included measures for overall trade flows (inflows + outflows), outflows (from country X to the USA or UK) and inflows (from the USA or UK to country X) (Barbieri et al. 2009). For reasons to be addressed in the discussion, we focus on log-transformations of trade outflows — flows from potential target countries to the USA and the UK.

We operationalize security linkages two ways. First, we include a measure of alliance patterns and foreign policy similarity (Signorino and Ritter 1999). This measure uses a combination of alliance membership patterns and United Nations voting data to produce a single measure, ranging from zero (no manifest interest commonality) to one (complete manifest interest commonality), on an annual basis. Second, we include arms transfers to potential target countries (log transformed). Both the USA and the UK are major arms exporters, with the USA accounting for 40 % of total sales, and the UK 7 %, to developing nations from 2002–2009 (Grimmett 2010). Because the variable measures arms transfers, rather than just sales, it captures the cumulative effect of sales, in-kind military aid, and aid that is tied to purchases of USA and/or UK military hardware. Data are from the Stockholm International Peace Research Institute Arms Transfers Database (SIPRI 2010).

3.3 Control variables

The most important controls are the extent of human rights abuses and population. We expect that AI would more often target countries with worse human rights records. To operationalize the prevalence of human rights abuses in a society, we use the Political Terror Scale (PTS) (Wood and Gibney 2010). The PTS is a widely used index that measures physical integrity rights violations, such as extrajudicial killing, torture or similar physical abuse, disappearances, and political imprisonment, committed by the state or its agents. The PTS provides codings based on AI’s annual report, The State of the World’s Human Rights, and on US State Department country reports. We use the AI-derived coding because it allows us to assess the effect of both domestic and international factors affecting AI’s targeting while controlling, as best as possible, for AI’s evaluation of human rights conditions in the country.¹³ This is the most direct, practical measure of AI’s assessment of a country’s “need” for human rights advocacy. The scale ranges from 5 (“Terror has expanded to the whole population. The leaders of these societies place no limits on the means or thoroughness with which they pursue personal or ideological goals”) to 1 (“Countries under a secure rule of law, people are not imprisoned for their views, and torture is rare or exceptional. Political murders are extremely rare”) (Wood and Gibney 2010, 373). We use the inverse of the PTS in our analyses, so that higher values represent greater respect for human rights.

Second, we control for population (log transformed), as we expect more populous countries to be targeted more often. Data are from Gleditsch (2002). There are several reasons why we expect this to be the case. First, regarding background reports and press releases, larger countries should provide more “bang for the buck”: a background report on Equatorial Guinea, with a population of just over one million, illuminates the adverse conditions of fewer people than a similar report targeting a larger country. Regarding UAs, larger countries, for any level of aggregate respect for human rights, will have more individuals that would be potential targets for individual abuse. We expect that these two variables should be highly significant, as they represent the most objective, non-political factors that would drive human rights reporting and advocacy: the severity of abuse and the number of abused.

We also include a host of additional controls. We control for regime type using the potential target country’s revised combined Polity score, commonly referred to as Polity2. Polity2 subtracts the Polity AUTOC score from the DEMOC score, producing a 21-point scale ranging from 10 (strong democracies) to −10 (strong autocracies). Because the effect of political democracy may be curvilinear – strong autocracies and strong democracies might display dynamics different from intermediate or hybrid regimes – we include the squared Polity2 term as well (Marshall et al. 2009). Ron et al. (2005) find that democracies are targeted for AI press releases more often, but that there is little or no effect of regime type on targeting for background reports.

We control for the incidence of civil conflict. Civil conflicts are dramatic, violent events that tend to draw media attention (Lacina 2004), and governments fighting internal wars have been demonstrated to perform significantly worse in respecting human rights, and as such may draw increased scrutiny (Poe and Tate 1994). We include a control for peace years, a count variable of the years a country has been at peace, as we expect that AI attention should decrease the longer a country experiences domestic peace. We include a control for past media coverage of human rights in a potential target country. Current targeting may be driven by media attention, and media attention is likely to be a significant influence on the salience of countries to AI members, IGOs, and powerful Western states. We use the average number of Newsweek and The Economist stories mentioning human rights in a country in a particular year, as coded by Ramos et al. (2007) and updated by Hafner-Burton (2008). Ramos et al.’s analysis demonstrates that it is a reasonable proxy for Western media attention.

We control for geographic proximity. Trade flows are highly negatively correlated with distance: countries tend to be more economically linked to their neighbors than to distant lands (Frankel and Romer 1999). Moreover, it is plausible that military linkages are more common with neighbors than with far-flung countries, and general country salience is likely a function of distance as well. Including the capital-city-to-capital-city distance to the USA and distance to UK will allow us to distinguish the effect of aid, trade and military linkages from more general proximity effects (Gleditsch and Ward 2001). We control for level of economic development with real GDP per capita (log-transformed) and include a time trend to account for a general increase in AI’s reporting and UA activity over the sample period (Gleditsch 2002). Finally, we control for whether a potential target country is involved in a militarized interstate dispute (MID) with the USA or UK. MIDs are “cases of conflict in which the threat, display or use of military force short of war by one member state is explicitly directed towards the government, official representatives, official forces, property, or territory of another state” (Ghosn et al. 2004).

3.4 Estimation

We run all models on a sample of 152 countries for the period 1977–2000. Because the distributions of the dependent variables are highly skewed and zero-inflated, we use zero-inflated negative binomial regression. This technique models the data generating process as two-step, in which the process that results in targeting versus no targeting – the zero observations – is potentially distinct from the process generating variation in the subsequent volume of targeting.¹⁴ We include the PTS score, log population, log GDP per capita, the Polity2 score, and the time trend in the first stage estimations. The PTS score and population are included for the same rationale that governs their inclusion in the second stage: they proxy objective “need.” Level of development and Polity2 are included based on their having emerged as significant predictors of scrutiny in earlier work (Ron et al. 2005).

In order to address cross-country heteroskedasticity, we estimate Huber-White robust standard errors clustered on countries. We estimate models with and without lagged dependent variables. Our theory is not truly dynamic, in the sense that reporting and/or UA targeting at time t is not necessarily a function of reporting and advocacy efforts at t-1 (Keele and Kelly 2006). However, all forms of AI and media reporting are strongly auto-correlated (r > 0.7); concerns regarding autocorrelation are warranted. Thus, we report estimates from both.

All explanatory variables are lagged one year in order to mitigate concerns about endogeneity arising from the potential that AI’s targeting of a country for advocacy is causing changes in the linkage variables. We acknowledge there is some potential for targeting to negatively affect the various linkage variables in the present and the future. The boomerang model is predicated on the assumption that Western states can be compelled to change their relations with potential violator states. In some cases, such as the campaign against South Africa during the anti-apartheid movement, NGO activists were successful in shaming Western governments to impose significant economic and military sanctions, i.e., linkage variables at time t were a function of advocacy at time t or t-n. Instrumental variable techniques may be preferable for identifying casual effects in situations where simultaneity is an issue. Given the number of theoretically important independent variables in the models (each of which would require a valid instrument), and that in most cases the most plausible instrument for present levels of trade, aid, or security linkages would be their lagged values, we did not employ instrumental techniques. Lagged indicators partially address this issue by accounting for temporal sequencing: present advocacy may affect future or present linkages, but present advocacy cannot affect past linkages.

The USA and the UK have very similar alliance patterns and UN voting records, trade patterns, and considerable overlap in the countries to which they donate ODA. This introduces problematic levels of multicollinearity into the models.¹⁵ Thus, we present models estimating the effects of US and UK linkages separately.

4 Results

Table 1 reports the results of our models using US linkage variables, while Table 2 reports those based on UK linkage variables. Briefly, the results of the first-stage estimates do not provide especially strong evidence of zero-inflation, or that targeting versus non-targeting is generated by a different data generating process than subsequent decisions about the overall volume of targeting. Across the US and UK specifications, past human rights performance matters, but in the opposite direction of that expected: countries with better human rights performance are more likely to be targeted for advocacy. Population is significant in some, but not all, of the models, indicating that smaller countries are more likely to be targeted. Finally, democratic regimes are more likely to be targeted for UAs, but regime type does not affect the likelihood of being targeted at all for background reports and press releases.

Table 1 US aid, trade, and security linkages and Amnesty International advocacy efforts

Table 2 UK aid, trade, and security linkages and Amnesty International advocacy efforts

We now turn to the second-stage results. Across all specifications, two variables that proxy need (in-country human rights conditions, population) are highly significant and in the expected directions: AI more frequently targets countries with worse human rights records and larger populations for all types of advocacy. In all lagged DV (LDV) models, the LDV is highly and positively significant. Past targeting is a significant predictor of future targeting. In addition, the time trend is positive and significant (p < 0.01) in all specifications. AI has increased its issuance of all types of advocacy over time, reflecting a general increase in organizational size and capacity. Looking at each type of advocacy in turn, however, we find different patterns with respect to the linkage variables.

Consistent with our hypothesis, background reports are not, in the main, correlated with aid, trade, and security linkages with the USA and the UK. Across the US and UK estimations, only log ODA is significant (p < 0.1) and only in a single model (Table 1, model 2).¹⁶ In addition to the controls proxying need, the controls for log GDP per capita (p < 0.01), civil conflict incidence (p < 0.05) and the time trend are significant across all specifications. Polity2 is negatively correlated with targeting for background reports in the US (p < 0.01) and UK (p < 0.1) LDV specification, indicating more democratic countries are targeted less frequently. Lagged media coverage is strongly significant (p < 0.01) and positive in the models without the LDV. Its insignificance in the LDV models suggests that media coverage and background reporting are jointly determined – AI reporting, along with press releases and UAs, drives media attention and vice versa. This pattern holds across the three types of advocacy. Two additional controls (MID with UK, log distance from capital to UK) are significant in the UK models.

We hypothesize that AI press releases would be correlated with aid and security linkages, which confer leverage. The results both confirm and counter our expectations. Arms transfers from the USA and foreign policy similarity with the USA are both highly significant and in the expected direction. Using model 3 as the baseline specification, a one standard deviation increase from the mean value for US foreign policy similarity (equivalent to going from China to Spain in 2000) is associated with a 19.3 % increase in the frequency of targeting for press releases, while a similar increase in arms transfers yields an 10.9 % increase. Both findings are consistent with our expectations. However, a third linkage that confers significant leverage potential – ODA – does not. ODA is negative in all four specifications and significant (p < 0.05, p < 0.1) in the UK models. Countries receiving more ODA from the UK are targeted for fewer press releases (−8.9 %, given a one standard deviation increase in log ODA from UK mean, based on Table 2, model 9). Regarding the controls, log GDP per capita is positively and significantly correlated with press releases in all four specifications. Polity2 is negatively and significantly correlated with press releases in the US models. While civil conflict incidence is not correlated with press releases,¹⁷ years since conflict are negatively correlated (p < 0.10, p < 0.05) with press releases in the US models, indicating that a tapering off of AI press releases follows conflict termination. Countries involved with MIDs with the UK are targeted much less frequently (p < 0.01, p < 0.1), while countries that are closer to the UK are targeted more (p < 0.05).

We hypothesized that all three types of linkages – aid, trade, and security linkages – would be associated with the frequency of targeting for UAs. This hypothesis receives substantial support – albeit only with respect to linkages with the USA. UK foreign policy similarity is negative and strongly significant in both UK specifications. None of the UK linkage coefficients are positively and significantly correlated with the frequency of targeting for UAs.

Trade from a potential target country to the USA is highly statistically (p < 0.01 in both models) and substantively significant in its effect: a one standard deviation increase from the mean value associated with a 28.7 % increase in the frequency of targeting.¹⁸ In additional specifications, both imports from the United States and the overall volume of trade (bilateral imports + exports) are less strongly correlated, in both the statistical and substantive senses, than trade from a potential target country to the USA. While trade matters for the targeting of UAs, aid does not. Coefficient estimates are small and statistically insignificant in all four specifications.¹⁹

The effects of security relationships with the USA are mixed: foreign policy similarity is strongly (p < 0.01 in both models) associated with targeting for UAs. A standard deviation increase in foreign policy similarity is associated with a 19.0 % increase in the frequency of targeting.²⁰ However, arms transfers are not statistically significant in any specification. Regarding the control variables, human rights conditions and population are both highly significant and in the expected direction. Log GDP per capita had a positive and significant effect across US and UK models.

There is a strongly significant, negative, and curvilinear effect of political democracy on the frequency of targeting for UAs (p < 0.01 in three of four models). The effect peaks when regimes are weakly authoritarian (Polity2 = −5), but is lower for the most authoritarian regimes and lowest for highly democratic regimes. Western media reporting is positive and significant only for the non-LDV specifications. Countries involved with MIDs with either the USA or the UK are targeted far less frequently. Finally, countries further from the UK see more targeting for UAs.

Taken together, these findings suggest three broad conclusions. First, violator countries’ international linkages with powerful Western states affect the volume of AI targeting for advocacy, and do so in predictable ways given the organization’s competing imperatives and the logics of information, leverage, and salience politics. These organizational imperatives at once encourage AI, similar to other human rights INGOs, to be impartial in its background reporting on abuses, but also disproportionately target countries with stronger security linkages to the USA for press releases, and stronger trade and security linkages for UAs. These findings are consistent with a theoretical model that highlights information and leverage politics – long-standing foci of the literature on transnational activism – and the concept of salience politics.

Second, AI’s advocacy efforts follow the US flag, but not the UK. While US linkage variables are often positive and significant in the various models, UK linkages are, in the main, not. Indeed, UK linkages are often negative, indicating that closer ties with the UK implied less targeting for advocacy. Since we control explicitly for AI-derived assessments of the prevalence of human rights abuses in potential target countries, the possibility that this is due entirely to selection effects – that the USA has ties with systematically worse abusers – is remote. Earlier research on AI reporting confirms a bias toward reporting on the USA and wealthier countries, but our findings suggest that this increased scrutiny extends to those countries that have stronger security and economic linkages with the USA. While an emphasis on states with stronger ties to the USA is consistent with the boomerang model of INGO advocacy, wherein INGOs press powerful Western governments to exert influence over their allies and aid recipients, the non-findings regarding UK linkages are not.

This conclusion raises the question of whether Latin American cases drive these findings. Previous research has found a significant pro-Latin America bias in both AI advocacy and media reporting on human rights abuses (Clark 2001, Hafner-Burton and Ron 2012). The Western hemisphere has long been a zone of strong US influence, and the early fame of AI was largely built around its rallying around human rights abuses in the region. Outside of Latin America and the Caribbean, UK and US security linkages tend to overlap. Is it the case that the divergent findings are driven by Latin American/Caribbean cases? Do our findings just capture “Latin American-ness”?

Table 3 reports the results of the US models with the addition of an indicator variable for the Latin American/Caribbean region. With respect to UAs, which require active member participation, we find weak evidence of a pro-Latin bias: controlling for the various types of aid, trade, and security linkages, AI targets Latin American and Caribbean countries for UAs more than twice as frequently (90.9 % more, model 17) as non-Latin American countries. However, the effect is only weakly significant (p < 0.1) in the LDV model, and not significant in the non-LDV model.²¹ This finding is consistent with several qualitative analyses of AI’s organizational and membership preferences, which confirm significant Latin bias (Clark 2001, Hopgood 2006). Moreover, once the indicator variable is introduced, only the trade linkage remains significant (p < 0.01 in both models).

Table 3 US aid, trade, and security linkages and Amnesty International advocacy efforts, with Latin America/Caribbean indicator

However, controlling for other domestic and linkage variables, the coefficients on the Latin America/Caribbean indicator variable are negative and insignificant for background reports and negative and significant (p < 0.05, p < 0.01) for press releases. US foreign policy similarity retains its statistical and substantive significance with respect to press releases, though the effect of arms transfers is rendered insignificant. For advocacy efforts emanating directly from the Secretariat, there is no Latin bias in background reporting and an anti-Latin bias in press releases: Latin American/Caribbean countries are targeted 51.5 % (model 15) less frequently than other countries, ceteris paribus. Given that human rights reporting by Western media disproportionately focuses on Latin American/Caribbean cases,²² this anti-Latin bias in press releases could be interpreted as a form of strategic substitution: AI may seek to focus media attention on regions and countries that are comparatively underreported, consistent with findings regarding INGO environmental naming and shaming (Murdie and Urpelainen forthcoming). Thus, a more nuanced picture of AI’s organizational preferences, and its purported Latin bias, emerges.

Third, AI does not shape its advocacy around all the variables that confer potential leverage. In particular, ODA – which should confer significant leverage potential – either exerts a weakly positive but not robust (the US models), or a suppressing effect (the UK models), on targeting for press releases. We take this as evidence of AI’s principled stand on development assistance. Whereas security relationships primarily affect the repressive capacity of the state, and therefore have direct bearings on state-initiated human rights violations, development aid ostensibly affects poverty alleviation. Since 2001 (but informally years earlier), AI has emphasized the “full spectrum” of human rights, which entail not just physical integrity and civil liberties, but also economic and social rights as well. The fact that AI chooses not to link ODA with leverage politics is inconsistent with a purely political-economic interpretation of its motives, and provides further evidence that AI balances its desire to affect human rights outcomes in target states with its ideological commitments. A similar logic helps explain why, in part, AI does not seek to use trade linkages as a potential source of leverage: economic sanctions hurt the very populations AI seeks to protect.

5 Extensions and conclusion

That organizations make choices is not a controversial point – we know from an ample literature in sociology and international relations that organizations have moral and material interests, and they act in ways to forward those interests. We have provided a framework for understanding how international linkages affect AI’s choices regarding targets and tactics for advocacy. Returning to the comparison of the Republic of Congo and Mexico in 1997, our models suggest that if Republic of Congo were to have had the same international linkages with the USA as does Mexico, it would have been targeted for nearly twice as many press releases (98.8 % more) and three times as many UAs.²³ This discrepancy is intriguing in light of recent research that shows naming and shaming to have positive, though often conditional, effects on human rights conditions in target states (Franklin 2008, Bell et al. 2012, Krain 2012, Hendrix and Wong 2013).

INGOs are principled and material actors engaged in the business of political change. As such, their targets for advocacy, and the methods of advocacy they choose, are both important. These factors dictate, to a certain degree, what we have come to know about human rights violations, and how we understand what it means to uphold human rights. INGOs are important conduits of information and engage in various activities to help frame the information they provide. They also act in strategic ways to convince states and IGOs to take action based on their reporting. We use the concepts of information, leverage and salience politics to highlight how INGO advocacy is shaped by the structure of international relations and the extent of aid, trade, and security linkages between violator governments and major Western powers.

Conventional understandings of what shapes human rights advocacy – such as the severity and extent of human rights abuse, population, and previous media attention – are in part confirmed, but are not helpful in explaining differences in how AI goes about its advocacy across cases. By using new data on UAs, we can ascertain differences in organizational choice among three types of advocacy: background reports, press releases, and grassroots letter-writing campaigns. Each of these three advocacy methods targets different primary audiences. All INGO advocacy tactics convey information, but at whom that information is targeted differs. This critical point is something that deserves closer examination in order to understand human rights INGOs as strategic and principled actors.

The fact that trade relationships and development assistance do not drive the targeting of press releases, AI’s principal form of engaging in leverage politics, demonstrates the importance of understanding AI as a principled, rather than purely strategic, actor. By contrast, security relationships, which are characterized by state-to-state transactions, provide a basis for AI to use leverage politics. If INGOs can get powerful Western states to alter their security relationships with violator states, they can influence violator governments by helping deny them access to weapons and funds that such states use to actively repress their populations. AI’s background reporting on human rights conditions largely conforms to the expectations of the information politics model. At least with respect to background reporting, AI appears relatively impartial.

More broadly, analyzing the logic of targeting also gives us a sense of how INGOs position themselves vis-à-vis other political actors: states, international organizations, and the media. Providing information is just one part of what INGOs and other non-state actors do. INGOs also decide at whom that information is targeted. Even though our current analysis relies on data about AI, this theoretical framework may be applicable to INGOs more broadly. Not all human rights INGOs have to answer to an active membership. For example, Human Rights Watch (HRW) does not have members. However, they have stakeholders – large donors, such as the Ford Foundation, as well as individuals, such as George Soros – who make demands on what HRW does and to a certain extent, can shape the types of advocacy campaigns the INGO pursues.²⁴ Other INGOs perceive a need to reach out to audiences beyond their traditional bases of support, forging ties to ordinary citizens, whether by staging spectacular shows of support in protest of certain policies (e.g., Live 8 concerts against poverty, organized by coalitions of INGOs, in 2005) or to increase their support base (e.g., Oxfam International’s expansion onto university campuses). In reaching these non-specialist and non-policymaking audiences, INGOs may be forced to rethink how to use the tools at their disposal, as detailed, impartial reporting might at times fail to generate as much interest as focusing on issues and cases that are already salient to the individuals/organizations whose support they are courting. As grassroots support becomes a more important source of INGO funding, we expect that INGOs will work to cultivate distinct niches, and that the issues and countries these supporters find relevant will be increasingly significant determinants of organizational behavior.

Going outside of INGO research, this argument ties in with other literatures that consider the choices that non-state actors make in order to influence outcomes. Barnett and Finnemore’s (2004) pioneering work on the principled and bureaucratic sources of IGOs’ decisions demonstrates clearly how internal priorities shape political outcomes. Others, such as Oestreich (2007) or Weaver (2008), have linked the dynamics identified by Barnett and Finnemore to the spread of human rights within different United Nations agencies and the difficulties faced by the World Bank in reforming organizational pathologies, tracing the interaction between internal organizational choices and external environments to create ideational change. The internal complexities of organizations lead to choices that cannot simply be attributed to either “rationality” or “principle” – rather, these logics interact.

Extant work on INGOs has focused largely on how INGOs have affected political outcomes. Alternatively, we explore how and why INGOs choose the targets and tactics they do. Instead of assuming either entirely political-economic or principled motivations, our analysis demonstrates that AI pursues multiple goals throughout its different tactics, and decisions to use different tactics are, in part, a function of relations between states. Largely as hypothesized, different types of linkages – whether aid, trade or security relationships – shape AI’s choice of targets and tactics. AI’s organizational prerogatives and the structure of international relations affect its advocacy efforts in systematic ways.