Communicating Elites

With regard to the elites engaging in communication, one of the hallmarks of global governance is the multitude of actors that aspire to influence its outcomes. Three types of elites are particularly prominent communicators about IOs: national governments, CSOs and the IOs themselves. National governments often criticize IOs in order to force policy change, deflect blame or mobilize domestic supporters, but may also endorse them to rally support for ambitious goals, lock in policy preferences and protect multilateral arenas (Morse and Keohane Reference Morse and Keohane2014). CSOs frequently challenge IOs for insufficiently ambitious policies and undemocratic decision-making procedures, but may also praise them for their policy achievements and efforts to consult with stakeholders (O'Brien et al. Reference O'Brien2000). Finally, IOs themselves increasingly engage in self-legitimation, trumpeting their democratic credentials, expertise-based policies and critical achievements, but also occasionally admit to mistakes when seeking to minimize political damage or generate support for reorientation (Zaum Reference Zaum2013).

There are reasons to expect that these elites exhibit varying degrees of credibility when communicating about IOs, which has implications for the effects of this communication on citizens' legitimacy perceptions. Credibility refers to the belief that a speaker has relevant knowledge and can be trusted to reveal that information accurately (Lupia Reference Lupia and Kuklinski2000). Prior research suggests that sources which are perceived to be more credible are more successful at shaping attitudes (Druckman Reference Druckman2001). According to this view, citizens are not mindless followers of elites who might want to manipulate them, but instead listen specifically to the elites they perceive to be credible on a particular issue. In the domestic context, partisanship has been found to constitute an important source of elite credibility: citizens are more likely to follow cues from party elites whose partisan orientation they share, especially in polarized political systems (Bullock Reference Bullock2011; Druckman, Peterson and Slothuus Reference Druckman, Peterson and Slothuus2013). In the global context, where elites are less strongly linked to specific partisan positions, and the issues are often less polarizing, we expect credibility to instead be tied to perceptions of impartiality. However, we also conduct robustness checks to test for moderating effects of partisan identification and trust in the domestic government on treatment effects.

This expectation of a link between credibility and impartiality is inspired by research on the impact of expert endorsements, which shows that experts can affect public opinion by virtue of their perceived unbiased knowledge (Guisinger and Saunders Reference Guisinger and Saunders2017; Maliniak et al. Reference Maliniak2019), as well as research on media priming, which shows that news sources perceived as authoritative are more likely to produce framing effects (Druckman Reference Druckman2001; Miller and Krosnick Reference Miller and Krosnick2000). Analogously, we expect that citizens consider whether elites in global governance can be expected to hold and reveal accurate information about IOs. Elites who have greater incentives to convey biased information about IOs are less likely to be seen as credible sources. Conversely, elites who stand to gain less from how IOs are perceived can be expected to communicate more honestly about these organizations.

Based on these considerations, we expect citizens to believe CSOs are more credible than national governments or IOs in their assessments of IOs for three reasons. First, CSOs are constitutively independent from IOs and are therefore more likely to be regarded as autonomous voices (cf. Gourevitch, Lake and Stein Reference Gourevitch, Lake and Stein2012). Many CSOs, such as Human Rights Watch and Transparency International, have made it their organizational purpose to offer independent assessments of norm conformance and goal achievement among IOs and their member states (Kelley and Simmons Reference Kelley and Simmons2015).

Secondly, national governments are the principals of IOs (Hawkins et al. Reference Hawkins, Lake, Nielson and Tierney2006). They have played a part in creating IOs, serve on their governing bodies and are primarily responsible for implementing their policies. As a result, governments often have particular views about how co-operation should develop, such as in conflicts over burden sharing in the EU, voting weight in the IMF and dispute settlement in the WTO. Because of these stakes, national governments are likely perceived as less credible communicators than CSOs.

Thirdly, IOs have the most far-reaching vested interests in debates about themselves. Their bureaucracies are committed to advancing the goals of these organizations, but depend on the support of their political environment to achieve them (Barnett and Finnemore Reference Barnett and Finnemore2004). We therefore expect IOs to be the least credible source of information about themselves. We find support for this gradation of credibility in data on popular confidence in different elites worldwide. Data from the most recent wave of the WVS (2010–2014) suggest that environmental and women's CSOs are more credible than both governments and IOs, and that governments are more credible than IOs such as the UN and NAFTA (see Appendix C). This leads to a first hypothesis:

Objects of Communication

We assume that elite communication about IOs typically attempts to affect individual attitudes by invoking two alternative grounds for endorsement or criticism: the procedures and performance of IOs. An extensive literature shows that favorable attitudes toward a political institution may be shaped by the procedures and performance of that institution – both in the context of domestic politics (Esaiasson, Gilljam and Persson Reference Esaiasson, Gilljam and Persson2012; Newton and Norris Reference Newton, Norris, Pharr and Putnam2000) and global governance (Bernauer and Gampfer Reference Bernauer and Gampfer2013; Dellmuth and Tallberg Reference Dellmuth and Tallberg2015). Thus elites interested in shaping citizens’ attitudes toward IOs frequently refer to these features. Procedural standards invoked in elite communication often relate to democratic aspects of IO policy making, such as inclusiveness and accountability, but can also pertain to expertise and efficiency (Binder and Heupel Reference Binder and Heupel2015). Performance standards include aspects of goal achievement, such as problem-solving effectiveness and collective welfare gains, but can also relate to the fairness of outcomes (Zürn Reference Zürn2018).

Whether citizens' legitimacy perceptions are most sensitive to the procedures or performance of IOs is a topic of debate. Procedural accounts submit that process criteria are the most important for people's perceptions of legitimacy. Even when institutions generate outcomes that are to their disadvantage, actors accept their exercise of authority because of how they were set up and operate (Hurd Reference Hurd2007, 71). Procedural accounts have an antecedent in Weber's (Reference Weber1978[1922]) notion of rational–legal legitimacy, which emphasizes properly administered rules by properly appointed authorities. The idea that legitimacy results from features of the decision-making process is prominent in contemporary theories of procedural fairness (Tyler Reference Tyler1990) and democratic legitimacy (Held and Koenig-Archibugi Reference Held and Koenig-Archibugi2005).

Performance accounts instead claim that legitimacy perceptions are determined by institutions' contributions to collective welfare and distributional outcomes. Substantive outcomes are considered more powerful in shaping the perceptions of institutions than the process by which those outcomes were produced (Hurd Reference Hurd2007, 67). This idea features prominently in the study of domestic institutions: ‘Government institutions that perform well are likely to elicit the confidence of citizens; those that perform badly or ineffectively generate feelings of distrust and low confidence’ (Newton and Norris Reference Newton, Norris, Pharr and Putnam2000, 61). In the global setting, it is commonly claimed that IOs historically have earned their legitimacy through the benefits they have produced for states and societies (Buchanan and Keohane Reference Buchanan and Keohane2006).

Cueing theory gives us no reason to expect that the effectiveness of elite communication would vary depending on the features of IOs that are invoked. If it is correct that citizens consider both the procedures and performance of IOs when forming legitimacy perceptions (Anderson, Bernauer and Kachi Reference Anderson, Bernauer and Kachi2019; Dellmuth, Scholte and Tallberg Reference Dellmuth, Scholte and Tallberg2019), then elite communication that invokes these features should be effective in both cases. This leads us to expect:

Valence

Finally, we consider how the tone of elite communication may influence the strength of cueing effects. As previously described, elite communication spans the full evaluative spectrum, from endorsing, praising and defending IOs to challenging, criticizing and dismissing them. All types of elites perform these positive and negative discursive strategies, which are frequently referred to as the legitimation and delegitimation of IOs, respectively (Binder and Heupel Reference Binder and Heupel2015; Tallberg, Bäckstrand and Scholte Reference Tallberg, Bäckstrand and Scholte2018; Zaum Reference Zaum2013; Zürn Reference Zürn2018). Empirical examples include US President Donald Trump calling NAFTA the worst trade deal the US ever signed (New York Times, 30 March 2017), Greek Finance Minister Yanis Varoufakis accusing the EU of terrorism during the country's economic crisis (BBC, 4 July 2015), and UN Secretary-General António Guterres praising the organization's peacekeepers for making the world safer (UN, 30 May 2017).

The study and practice of legitimation and delegitimation assume that the evaluative component of communication matters. It is by praising or criticizing IOs that elite messages become potentially powerful in shaping citizen attitudes. Few would expect neutral elite messages to be influential. This expectation is borne out in studies showing that positive and negative elite cues shape public support for the EU (Maier, Adam and Maier Reference Maier, Adam and Maier2012) and on international issues generally (Guisinger and Saunders Reference Guisinger and Saunders2017). However, it is an unexplored question whether legitimation or delegitimation is systematically more or less effective at shaping citizen attitudes toward IOs.

We expect negative elite messages to have stronger effects on legitimacy perceptions than positive messages. We base this expectation on theories in comparative politics, economics and psychology. While identifying slightly different mechanisms, all ground their expectations in general socio-psychological dynamics, and all suggest that negative messages should have a larger impact.

Research on voting behavior shows that people respond asymmetrically to positive and negative information about the economy (Bloom and Price Reference Bloom and Price1975; Soroka Reference Soroka2006). Since people tend to be slightly optimistic in their basic predisposition, negative information usually diverges more from their reference points than positive information, and therefore has a greater impact on attitudes and behavior. This dynamic has also been identified in communication about political candidates and institutions (for example, Lau Reference Lau1985).

Prospect theory suggests a similar story of asymmetry, highlighting a complementary mechanism. It submits that individuals tend to be risk averse, and weigh potential losses more heavily than potential gains (Kahneman and Tversky Reference Kahneman and Tversky1979; Tversky and Kahneman Reference Tversky and Kahneman1981). Such loss aversion leads people to react more strongly to negative information than to positive, for instance, by cutting back consumption more sharply in response to bad news than they expand consumption in response to good news.

Finally, psychological research on impression formation establishes that bad emotions weigh more heavily than good emotions. Negative information tends to be processed more thoroughly, be stickier and have a greater impact than positive information. As Baumeister et al. (Reference Baumeister2001, 323) summarize: ‘The greater power of bad events over good ones is found in everyday events, major life events, close relationship outcomes, social network patterns, interpersonal interactions, and learning processes…Bad is stronger than good, as a general principle across a broad range of psychological phenomena’.

We expect these general socio-psychological dynamics to also influence how people respond to communication about IOs. They suggest a third hypothesis:

Survey Design

We conducted our survey experiment among nationally representative samples in three countries – Germany, the UK and the US – in order to reduce the risk of biases from contextual country factors. Averaging the communication effects across populations from different countries strengthens the external validity of our findings. We selected these countries since they: (a) are politically central in the examined IOs, which makes our findings substantively important for the prospects of global governance; (b) are democratic, which avoids the issue that the legitimacy of political institutions may mean different things to citizens of democratic and autocratic regimes (Jamal and Nooruddin Reference Jamal and Nooruddin2010) and (c) have very high levels of Internet penetration (over 80 per cent), which increases our confidence in the external validity of the data.

To implement the questionnaire, we relied on online panels from YouGov, a well-reputed global survey company (Berinsky, Huber and Lenz Reference Berinsky, Huber and Lenz2012). YouGov uses targeted quota sampling with the aim of achieving representative samples at the end of the fieldwork. The samples for our survey were matched to the full populations of the three countries using age, education, gender and party identification. To enable researchers to take national differences in sampling procedures into account, YouGov creates a variable denoting the optimal weight that should be assigned to each observation in order to achieve nationally representative results (for a detailed discussion, see Ansolabehere and Schaffner Reference Ansolabehere and Schaffner2014). While the sample should thus be generalizable to the populations of these countries, the YouGov online panel is self-recruited, which may introduce motivational factors. Participants receive small monetary incentives for their participation, such as entries into prize draws. Since our goal is to establish causal effects through an experiment with high internal validity, rather than to identify absolute levels of perceived legitimacy in a population, we consider this risk acceptable. A total of 3,270 interviews were conducted in the UK, 3,268 in Germany, and 3,135 in the US in January 2015. In addition to the experiment, the survey (Appendix A) included several attitudinal and demographic questions, which we use to describe the country-specific samples (Appendix Table B1), and for balance tests (see below).

Experimental Conditions

To isolate the causal effects of elite communication, we randomly assigned individuals to groups that received different experimental treatments, and a control group that did not receive any treatment. We used a vignette approach to treatment, which is well suited for complex factorial experiments (Mutz Reference Mutz2011, 54). Vignettes are short statements that contain the treatment and precede the question of interest. The purpose of vignette treatments is to assess what difference it makes when the factors embedded in the vignette are systematically varied. In this case, we manipulated three features of the communicative situation: the elite making the statement (Hypothesis 1), the object of communication (Hypothesis 2) and the valence of the message (Hypothesis 3). The vignettes were formulated in a way that allowed us to vary the three factors with precision, but also to express the subject matter in concrete terms so that it would be understandable to respondents (Gibson Reference Gibson2008). Moreover, we sought to formulate vignettes that would work equally well for all IOs and that were short and straightforward, since longer and more complex vignettes make it more difficult to determine what individuals respond to (Mutz Reference Mutz2011, 64–65).Footnote ²

The treatments combined into a 3 × 2 × 2 factorial design, with a total of twelve conditions (Table 1). We allocated the same number of individuals to each combination of factors, and the number of respondents who provided a substantive answer was eventually relatively even across groups (see Appendix Table B2).

Table 1. Vignettes

To examine Hypothesis 1, we varied the elite making the statement in the vignette: CSOs, national governments or IOs themselves. To assess Hypothesis 2, we formulated vignettes about the procedures and performance of IOs: procedural vignettes invoked IOs’ democratic character and performance vignettes the problem-solving effectiveness of IOs.Footnote ³ To evaluate Hypothesis 3, we designed the vignettes so that they included positive or negative statements about IOs.

We operationalize legitimacy perceptions by using a question about confidence in IOs: ‘How much confidence do you personally have in the [IO] on a scale from 0 (no confidence) to 10 (complete confidence)?’ While legitimacy perceptions are a complex phenomenon, confidence has three main advantages as an indicator. First, it aligns well with our conceptualization of legitimacy as the belief that an institution exercises its authority appropriately. The confidence measure taps into respondents' general faith in an institution, independent of short-term satisfaction with specific policy decisions and outcomes (Easton Reference Easton1975; Norris Reference Norris2011). Secondly, a narrow measure of legitimacy beliefs, such as confidence (or trust), has advantages when studying the sources or effects of legitimacy. Different from some alternative operationalizations (for example, Esaiasson, Gilljam and Persson Reference Esaiasson, Gilljam and Persson2012), confidence does not integrate into the measure either potential institutional sources of legitimacy (such as the perceived fairness or effectiveness of an institution) or potential consequences of legitimacy (such as compliance with an institution's rules). While some studies use a multi-item measure to capture various complexities of legitimacy as a concept, these studies usually invoke a broader conceptualization of legitimacy, incorporating normative standards to be met by an institution and/or acceptance of the rules of an institution (for example, Anderson, Bernauer and Kachi Reference Anderson, Bernauer and Kachi2019; Gilley Reference Gilley2006). Thirdly, the confidence measure allows us to relate the findings of this study to the large literature on public opinion that also employs confidence (or trust) as an indicator of legitimacy perceptions (for example, Bühlmann and Kunz Reference Bühlmann and Kunz2011; Dellmuth and Tallberg Reference Dellmuth and Tallberg2015; Dellmuth and Schlipphak Reference Johnson2019; Johnson Reference Johnson2011; Norris Reference Norris2011; Voeten Reference Voeten2013).

In the experiment, the control group only received the question about confidence in a particular IO. The treatment groups (Table 1) received a specific vignette first and then the confidence question. Respondents were never allocated to the same treatment group twice. Respondents who were placed in the control group remained in this group throughout all four rounds.

Experimental Rounds

Moving beyond the strict hypothesis test, we also explored the extent to which communication effects vary across IOs. We therefore conducted the survey experiment in several rounds; each round performed the same experiment on a different IO. We selected five IOs that are central in their respective policy domains and prominent in public debate: three at the global level (IMF, UN and WTO) and two at the regional level (EU and NAFTA). While some IOs fly beneath the radar of public awareness, we selected organizations that are both known to citizens at a basic level and regularly subject to positive and negative communication by elites.Footnote ⁴ This ensures that treatments expressing elite messages about these IOs are understandable and reasonable to respondents. At the same time, the levels of citizen familiarity and public debate differ across these IOs, suggesting potential explanations of variation in the treatment effects, which are further explored in the comparative analysis. Respondents were asked about the regional IO of which their country is a member. That is, the question about confidence in NAFTA was only asked in the US, and the question about confidence in the EU only in the UK and Germany. The order of the experimental rounds for all respondents was: UN, EU/NAFTA, IMF and WTO.Footnote ⁵

Robustness Checks

We perform seven robustness checks. First, we replicate all analyses presented in Figures 1–6, including country dummies to check whether some of the results are driven by potentially unobserved country-specific variables. This change in model specification does not alter the interpretation of our results (see Appendix F).Footnote ⁹

Secondly, we go one step further and replicate all analyses at the level of individual countries (Appendix G). The main results from the pooled analysis are robust across countries with a few notable exceptions. In Germany, the varying effectiveness of negative and positive treatments is compounded by positive communication not swaying citizen confidence in IOs at all. The expected results for communicating elites, objects of communication and valence remain robust (Tables G1–G3). In the UK, the results hold across the board, except that negative communication by CSOs is not effective and the government does not appear to be more effective at swaying public opinion than IOs themselves (Tables G5–G7). The latter finding may be due to the extreme dissatisfaction with the British government at the time of our survey. According to the British Social Attitudes survey, only 17 per cent of the population had confidence in the government in 2015.Footnote ¹⁰ The main results also hold in the US, even if fewer treatments are significant (Tables G9–G11). The main deviation is that the government appears to be a more effective communicator than both CSOs and IOs themselves (Tables G9). This may reflect greater skepticism about the impartiality of CSOs in the US than elsewhere (Gourevitch, Lake and Stein Reference Gourevitch, Lake and Stein2012). When we disaggregate the country analyses by IO, the pattern is somewhat more heterogeneous, likely reflecting a combination of country-specific experiences of IOs and less statistical power (Tables G4, G8 and G12).

Thirdly, we examined whether eight different individual characteristics measured in the survey, including education and age, are evenly distributed across the conditions we aggregated for the analysis. The results increase our confidence in the randomization of the subjects among treatment groups. We only discover imbalances in six of the ninety-six tests. Balance tests using the twelve treatment conditions in Table 1 show four imbalances for ninety-six tests (see Table 1 and Appendix Tables H1–H2).

Fourthly, we examine whether there are potentially undesired spillover effects because the order of the four experimental rounds was not randomized. For this purpose, we conducted balance tests for each round separately to assess if the fixed order of the rounds could have given rise to biases resulting from dropouts (Appendix Tables H3–H10). We found no pattern indicating a potential systematic bias, as approximately the same (very low) number of balance tests is statistically significant in each round. Still, we examine potential biases further, as the absence of randomization of experimental rounds may give rise to varying distributions of respondents across samples. Indeed, the number of respondents providing substantive answers drops when comparing rounds 1 and 4 (Appendix Table B2). We test whether the experimental effects vary across the four rounds by plotting the predicted marginal effects of the various treatments for different rounds (Appendix I). The slopes of the dummy variables for the specific treatments are largely parallel across the four experimental rounds, indicating an absence of systematic differences.

Fifthly, we replicated all analyses in order to check whether item non-response may have affected the results. If the use of the ‘don't know’ option reflects, for instance, less political knowledge about IOs, these values would not be missing at random (Rubin Reference Rubin1976, 582) and average treatment effects may be biased. We therefore examined the causal process behind missingness, and found that item non-response is unlikely to have affected our results (see Appendix J). While the effects of socio-demographic factors become smaller when comparing rounds 1 and 4, possibly because the samples have become more homogenous due to increasing item non-response, we also find instances of effect sizes becoming larger when comparing across other rounds (for example, education across rounds 3 and 4, and age and gender across rounds 1 and 2). Taken together, the evidence from the balance tests, plots of experimental effects across the four rounds, and item non-response analysis corroborates our confidence in our interpretation of experimental effects across rounds in light of our theoretical argument.

Sixthly, we examine whether the results for governments as communicating elites are conditional on whether people identify with a political party in government and whether people trust their own government. We find that, while the effect of the positive government treatment is unconditional on these two factors, the effect of the negative government treatment is moderated by partisan identification and confidence in national government (Appendix K). More specifically, the negative effect of the negative government treatment is only statistically significant among those who identify with a party in government (Appendix Figure K1) and becomes stronger the higher the respondent's trust in government (Appendix Figure K2). These findings are in line with previous research suggesting that people who distrust their own government are unlikely to follow government cues (Aaroe Reference Aaroe2012), and that a political party's cues work best among those who identify with that party (Maier, Adam and Maier Reference Maier, Adam and Maier2012).

Finally, we examine whether treatment effects might be conditional on political awareness. A large public opinion literature suggests that politically aware individuals are more likely to comprehend and integrate new information into their opinion formation (Druckman and Nelson Reference Druckman and Nelson2003). Yet they also tend to hold more consistent and stable opinions, and may therefore be less responsive to experimental treatments (cf. Zaller Reference Zaller1992). We test these issues by interacting the treatment dummies with two awareness indicators (Appendix Tables K3–K8). The results suggest that more knowledgeable citizens did not respond differently than less knowledgeable citizens when confronted with our vignettes.