Abstract
As a populist leader, Hugo Chavez famously used emotionally charged populist rhetoric to appeal to a broad base of poor and working-class Venezuelans. Was his choice of linguistic discourse a tool of popular control, response to public opinion, or both? Answering this question sheds light on the effectiveness of classical democratic conceptions of vertical accountability for populist leaders. Using a theoretical framework incorporating macro implications of Zaller’s receive-accept-sample (RAS) model, the concept of Erikson, Mackuen, and Stimson’s mood, and latent public opinion, we develop several competing expectations regarding rhetoric and presidential approval in Venezuela. Using computational sentiment analysis on a unique dataset of transcripts from Chavez’s Aló Presidente broadcasts, we evaluate Chavez’s quarterly public approval ratings with vector autoregression (VAR) and Koyck models. Results indicate presidential approval levels are causally linked to not only exogenous economic factors but also leader discourse. Results also indicate that leader language is not shaped by approval levels, illustrating the power of messaging and media control for populist leaders and the potential limits of democratic accountability.
The ability for the electorate to hold elites accountable via vertical accountability is at the heart of normative and positive democratic theory. 1 A popular formulation of democratic theory argues that elites respond to public attitudes and opinions while the public’s opinion is shaped by elite performance. This view has been challenged many times, mostly along two key dimensions. The first is the argument that elites have significant influence over opinion formation among key sectors of the public. As Zaller (1992) argues, elites often control issue messaging, providing a powerful tool for influencing opinion formation. This line of argument has further been substantiated in differing periods in the United States, Europe, and in some developing contexts (e.g., Druckman 2001a, 2001b, 2001c; Geddes and Zaller 1989; Druckman and Lupia 2000; Stockmann and Gallagher 2011). Although much of Zaller’s model is focused on the individual level, its implication in the aggregate is of a central, if restricted, role for elites in shaping public opinion across a range of issues.
A second challenge to the classical model focuses on events out of the control of leaders. This literature argues that the public rewards and punishes leaders for events over which they have no control—from global economic markets, to natural disasters, to shark attacks, or basketball games (Achen and Bartels 2016; Campello and Zucco 2016; Healy, Malhotra, and Mo 2010; Hellwig 2014). The link between elite performance and approval may be tenuous at best. This weak linkage is possibly even weaker in the context of populist leadership where charisma and leader discourse is at the heart of electoral success, such as Hugo Chavez’s tenure in Venezuela.
In democracies, these two views of vertical accountability (or lack of it) are in tension with each other. If leaders can control messaging and strongly shape public opinion, we should not see much of an impact from exogenous events. For semi-democratic regimes where competing messages are constrained, elite-driven narratives should exert more influence on public opinion as avenues of media access are, in part, controlled by the state. Geddes and Zaller’s (1989) work illustrates the efficacy of an elite-cueing model in authoritarian Brazil’s restricted media environment. If one of these approaches to accountability effectively works in populist regimes, its effect could be substantial during this period where populists appear to be ascendant.
The paper proceeds as follows. First, we highlight the claims of leader and mass-based theories of leader support along with a review of the extant literature on executive accountability and leader discourse, noting its limits regarding theoretical development and empirical evidence, particularly for populist regimes such as Chavez’s Venezuela. Second, we introduce the case of Venezuela in detail. Then we introduce two unique datasets, a quantitative analysis of Chavez’s language from his weekly television broadcast, and the Venezuelan component of the Executive Approval Dataset, a quarterly measure of Chavez’s public approval. The next section empirically examines relationships between Chavez’s discourse and his approval. The paper closes with a discussion of the normative implications for executive accountability in democracies led by populist presidents.
Who Leads? Accountability in Electoral Democracies
In electoral democracies, elections are the chief mechanisms of vertical accountability. Through the ballot box, citizens can compel leaders to take voters’ preferences seriously by allowing citizens to punish or reward leaders for their promises and performances. But accountability does not vanish between executive elections. Instead presidents use the public’s evaluations of their performance as windows into citizens’ evolving views and, thus, as a guide to adjusting their public discourse, rhetoric, and policy program. Although the effectiveness of public opinion as an instrument of vertical accountability is limited because negative evaluations do not automatically lead to the ouster of the incumbent, changes in presidents’ polling numbers are consequential enough to give presidents strong incentives to pay attention—and they do. 2 High polling numbers may facilitate the passage of legislation and increase party cohesion while low polling numbers can embolden opposition legislators and frustrate the president’s policy agenda. This is likely to be even more important in populist systems with weak parties, as for a populist leader holding office is likely less a function of structural party support than personal appeal. The populist leader who loses the ability to rally mass support invites rivals from across the political spectrum. At the far end of the spectrum of punishment for poor popularity is early removal from office due to deep and persistent declines in approval. As the recent impeachment of Dilma Rousseff in Brazil highlights, Latin America has no shortage of such examples (Pérez-Liñán 2007).
Research on Latin America has also produced mixed results regarding the role of the economy on vote choice. Although Cuzan and Bundrick (1997) find some links between economic conditions and presidential approval in Central America, Johnson and Schwindt-Bayer (2009) studying the same countries (but with a longer time series) find a consistent negative effect of inflation on presidential approval and no effect for growth. This mixture of results points to case heterogeneity and the role of leadership style; limits of the classical model of vertical accountability may be reached within the contexts of Latin American countries.
Understanding the factors shaping executive-citizen linkages between elections (i.e., approval) in new democracies is a relatively new endeavor as comparative public opinion data sources have developed over the past two decades. Taking advantage of the increasingly rich data, a small yet growing literature looks to understand factors that condition approval (Carlin, Love, and Martínez-Gallardo 2015a, 2015b; Powell and Whitten 1993). However, within this literature, there is a dearth of work on the impact of leader language, behavior, and discourse on approval ratings. In particular, we have a sparse understanding of how leader discourse can systematically shape approval directly or by mediating the effects of policy performance. In cases like populist Venezuela, where government controls much or all of the public media, leaders have an advantage in crafting and disseminating their messages to a captive audience. Chavez’s prolific televised oratories provide ample empirical data.
Although there is less written about leader discourse and executive approval, the literature on elite messaging and individual political attitudes and leader actions is more fully developed. This body of knowledge provides a foundation for a theoretical framework to understand the role leader language can have in popular control. This literature, as exemplified by Zaller (1992), Moe (1981), and Erikson, Mackuen, and Stimson (2002), among others, centers on how information flows from elites to the masses (and vice versa) in response to political and social events and conditions.
A large segment of the extant literature on elite messaging and popular control builds on Zaller’s elite-based receive-accept-sample (RAS) model. In it, elites, and elite discourse, play a key role in reinforcing partisanship and support for specific actors because politically engaged citizens (likely voters) tend to only receive and accept (the R and A in RAS) reinforcing messages. Elites potentially have immense power over shifting the policy focus and option sets of their political base. For a populist leader like Chavez, the RAS models highlight the potential extensive ability leader messaging and rhetorical control has to shape leader support.
The RAS approach also highlights the importance of message credibility for message acceptance. That is, voters consider the credibility of the message and messenger when evaluating elite discourse. This does not imply that the message is either truthful or accurate, only that the audience believes the speaker and the message. Often this takes the form of partisan or ideological alignment; however, in a populist regime with a failed party system, a leader like Chavez holds distinct advantages for increasing credibility by having a fractured opposition (limiting partisan choices and competing issue salience) and media control (limiting opposing information).
Although the microfoundations of the RAS model are at the individual level, and it was initially conceived to explain variance in survey response, it holds substantial macro implications for populations as a whole. As suggested above, the cueing ability of elites in Zaller’s framework provides a set of tools available to politicians for shaping individuals’ attitudes, even if the effects are limited partially to their core constituencies. In the case, Zaller examines—the United States—there are key competing elite cues from the mass media, potentially producing conflicting and competing cues. But with elected populist leaders, such as Chavez, these constituencies are rather large and messaging opportunities plentiful. Chavez controlled a substantial amount of media access, limiting elite cues to one group, one message. At times during Chavez’s tenure, some media outlets critical of the regime at times were shut down, limiting conflicting and competing messaging. This simplification of limited message competition makes aggregate predictions from the RAS model clearer and stronger. Thus, the logic of the RAS model implies populist semi-democratic leaders have the ability to shape their own level of public support through the careful use of message language and cueing.
Although the RAS model points to populist leaders having substantial opportunities to shape and manipulate their approval numbers, there is a competing literature that views the public as a more powerful independent agent. The work by Erikson, Mackuen, and Stimson (2002) and the related literature that Macro Polity spawned indicates the public may not be so easily swayed (Dalton 2013; Jones and Baumgartner 2005; Nicholson-Crotty, Peterson, and Ramirez 2009). As Erikson, Mackuen, and Stimson (2002) demonstrate for the United States, it is possible for relatively low information voters in the aggregate to demonstrate sophisticated vertical accountability for policy performance and ideology. They show support for the executive moves with political events and economic performance, and this pattern is predictable and regular. As with a large and liquid market, individual voters need not be well informed and rational for aggregate behavior to appear so. In addition, they develop and test the theoretical contention that there is a public “mood” regarding acceptable policy space. Although elites may shape the “mood,” they are also very much beholden to it. In their framework, elites’ ability to shape public support through messaging is limited by the public’s “mood” for a given policy area.
The concept of “mood” in some ways mirrors Key’s (1961) concept of “latent public opinion.” Deep-seated public opinion is likely to shape electoral choices and less apt for media and elite manipulation. This may lead to elites anticipating the public’s “mood” and subsequent likely shifts in opinion. Zaller (1992) raises this as a possibility and in later work (Zaller 1998, 2003) argues this likely shapes the public’s ability to independently and effectively influence elite actions. But this may be an overstatement. Politicians may be able to anticipate and preempt public opinion or “mood.” That is, the relationship between “mood” and elite discourse could be a dynamic one.
For the question of vertical accountability within a populist political context, we conceive of the idea of “mood” in a general conception for mapping the theory onto the case of populist leadership. That is, policy “mood” in the United States context, which is placed on a left-right socioeconomic scale, can be mapped onto a similar concept regarding approval of a populist leader: Chavez. Approving or disapproving of Chavez places the public along a dimension in support of populist-leftist leadership. In addition, being pro-Chavez involves both support for leftist concepts and populist leadership. In the case of Venezuela, these overlap substantially because the country lacks a credible recent history of rightist populism, as does much of Latin America. Thus, our implementation, and testing, of mood’s theoretical concept is one directly linked to the Macro polity framework: support for political leaders is linked to general support for their policy stances and leadership style.
Three key factors in Venezuela allowed a populist leader to overcome the informational challenges of anticipating, and potentially shaping, public “mood.” First is Chavez’s classic and well-honed populist tool of direct communication to the electorate through presidential addresses and his TV show. 3 Through these mediums, Chavez could avoid some of the message filtering conducted at the level of the media as well as political parties. The second factor is state influence of media, particularly broadcast media, in Venezuela during much of Chavez’s tenure. This allowed him to exercise greater control over the timing and content of political messaging received by the public. In combination, these factors restricted the informational environment for the public while increasing government control of the messaging. The final factor is the country’s reliance on oil exports for government spending (around 70% of the government’s budget comes from oil revenues), key resources for funding political patronage and populist policies. Because the international price of oil had a large impact on the resources available to Chavez along with Venezuela’s overall economic health, Chavez had a tool (oil spot and futures markets) for predicting future (in the medium- to short-term) challenges and windfalls often unavailable to political leaders.
The literature on elite messaging and language points to two competing outcomes. One outcome is where the public, via “mood” or latent opinion, shape leader actions and messages. The other is a political world where a leader can effectively use language and messaging to shape public opinion in their favor. But much of this theoretical and empirical debate regarding the efficacy and nature of vertical accountability thus far has focused on the United States and other advanced industrial democracies. These cases have well-established and stable party systems facilitating vertical accountability along both classical lines (i.e., public “mood” and sentiment) along with the cuing and messaging mechanism of the RAS model. What is much less well understood is how executive accountability operates in a political environment where discourse is populist in nature, traditional parties have failed, and economic conditions are directly and visibly linked to global markets (Campello and Zucco 2016). Such less developed, resource-based economies with weak party systems are the norm in Latin America and other developing regions, constituting a growing portion of the world’s electoral democracies. In particular, the rise of populist leaders in Latin America has further called into question the traditional conceptions of the relationship between leaders and masses when it comes to accountability and discourse. This paper tests the contention that vertical accountability in a poorly structured, yet contentious and populist political environment, favors elite control of voter attitudes versus vertical accountability derived from leaders responding to public “mood.” To put it succinctly, does Chavez’s language shape approval or does it move in response to approval and political realities (economic performance)?
The Archetype Modern Latin American Populist: Chavez
The particular populist context in which we test these contentions is Hugo Chavez’s Venezuela, an exemplar case of an emergent populist regime developed from a failed party system that characterizes a range of developing electoral democracies. Few leaders in Latin America have been as closely aligned with the practice of populist politics in the twenty-first century as the late Venezuelan president. Throughout his fourteen-year term, Chavez regularly and famously alternated between fiery rhetoric against his foreign and domestic opponents, and personal, uplifting stories to rally his supporters. Hawkins (2010) notes Hugo Chavez’s speeches display some of the highest levels of populist discourse in Latin America. Popular commentators and academics alike attributed Chavez’s substantial political success not only to Venezuela’s oil industry but also to his carefully constructed populist appeal—in both its rhetoric and action. Chavez’s language and strategies mirror Weyland’s (2001, 5) description of populism: “A charismatic individual wins and exercises power by maintaining direct, unmediated contact to a largely unorganized mass of followers.” This paper also views populism as “an ideology that considers society to be ultimately separated into two homogeneous and antagonist groups, the ‘pure people’ versus the ‘corrupt elites’” that uses “a highly emotional and simplistic discourse directed at the ‘gut of the people’” (Mudde 2004, 542). Following in this tradition, we see the populism of Chavez to be a thin ideology of the people versus elites (economic and political) conveyed by an appeal that is based on personality, charisma, and discourse of the movement leader: Chavez.
Throughout his nearly four terms in office, Chavez was a constant fixture on Venezuelan television, with government officials boasting at times that he is on television an average of forty hours per week (Hudson 2007). The centerpiece of this media strategy was Chavez’s weekly television show on the state-owned television network, Aló Presidente. Every Sunday at 11:00 a.m., Chavez hosted a mostly unscripted talk show which aired uninterrupted for four to eight hours with no fixed timeslot, ending at Chavez’s discretion. Chavez had regular access to the living rooms of his constituents, as well as to his cabinet of ministers, who were literally a captive audience as they were required to attend the programs in person. Chavez engaged with his audience in forums outside of Aló Presidente as well, including Twitter, interviews, televised speeches, and written communiqués. Although the entire nation may not have tuned in each and every week to watch Aló Presidente, the program represented a significant intrusion into Venezuelan citizens’ political lives.
One of the key questions facing those who study the behavior of individual leaders is how a case study–type approach can facilitate the development of generalizable claims about other leaders and the political spaces they occupy. Scholars have recently probed the role of individual leaders in international politics (Dyson and Preston 2006; Hermann et al. 2001; Jervis 2013). Jervis (2013) demonstrates how individual leader traits matter to foreign policy decisions but also how democratic institutions shape and constrain their options and, moreover, how exogenous events and external forces contextualize their tenures in office. Conversely, Hermann et al. (2001) argue the leader is the authoritative decision-making unit, and leaders can be pragmatists about foreign policy options, or crusading opportunists. The use of populist language is not exclusive to leftist populists or populists in general, as most, if not all, political leaders deploy this type of rhetorical strategy from time to time. However, for leaders like Chavez—whose presidency is fundamentally characterized as populist in nature (Hawkins 2010, 206; Mudde and Kaltwasser 2012)—this framework for evaluating leader language and popular support is useful. The purpose of focusing on populism and charisma for case selection is not to say that the anger and optimistic language is only the provenance of populist leaders; rather, we view the case selection as a “hoop test” in the conception of Bennett (2010). If leader language is a tool for mass control, or is obviously responsive to mass opinion (and not just elite audiences), then it should be most clearly observed in a populist context. If language appears not to be an effect tool of mass control, the findings would argue for a crucial reexamination of the role of populist discourse, or any leader discourse, in mass control and political success.
Chavez served as an archetype for Latin American populism as well as followed in the long populist tradition of leaders such as Juan Peron of Argentina. Recent leaders such as Ecuador’s Correa, Bolivia’s Morales, and Nicaragua’s Ortega have all emulated to some extent Chavez’s methods. Examining how Chavez’s language, and changes in key political events, shapes executive approval gives us insight into the functioning of a key regime type: leftist-populism. This is particularly pertinent as nearly all of Latin America is ruled by electoral democracies and populism has become a useful tool of persuasion and control for elites from Argentina to Mexico.
More broadly, populism, in both its leftist and rightist incarnations, is an emerging trend in mature democracies across Europe and North America (Mudde 2007; Mudde and Kaltwasser 2012) where political candidates and leaders use populist rhetoric to unify citizens against such issues as regional and global integration, deindustrialization, and immigration. Resurgent populism is a geopolitical force that extends the potential for generalizing the findings presented far beyond Latin America, as evidenced by Donald Trump’s election, the 2016 Brexit vote, Marine Le Pen’s candidacy in France, or the rise of the Danish Peoples’ Party—all emblematic of a larger populist phenomenon emerging in the developed world.
Language and Populism
Chavez leveraged the support of the people to buoy his presidency, and his link to them was likely through populist language and rhetoric. As Hawkins (2010) illustrates, the general language profile of Chavez’s presidency, and even policy framework, is illustrative of populist governance. If one of Chavez’s keys to political success was the nature and structure of his discourse, then we should expect linkages between language, approval, and economic resources.
Kazin (1998, 1) notes that purveyors of populist rhetoric see “ordinary people as a noble assemblage not narrowly bounded by class, view their elite opponents as self-serving and undemocratic, and seek to mobilize the former against the latter.” He further emphasizes that populism leverages rhetorical optimism, encouraging ordinary citizens and promoting the notion of social mobility through hard work. Populist rhetoric simultaneously seeks to coalesce a collective nationalistic identity, while also elevating the status of the leader as the champion of oppressed peoples. For example, in a February 2006 episode of Aló Presidente, Chavez said the following: All of life is working and poverty for all of life, it is slavery! It is another form of slavery, that is, one becomes a slave to work, live only to work and have some half-eaten food to give the children, women: that is not life, we want to free them all and end feudalism, new types of slavery are gone . . .
4
This “working poor” narrative connects to a broad audience while needling and challenging the elites. Over the course of Chavez’s tenure, his ideological perspective appeared to shift from nationalism to one with a more socialist, at times Marxist, tone (Brading 2012; Hawkins 2010). Although this shift appears to have occurred around 2006, in connection with increased anti-American sentiment, we do not find evidence that it affected Chavez’s language markers examined below, nor does it appear to have caused a marked shift in his approval. 5
The strategies Chavez used, and outlined above, can be explained in part by the elaboration likelihood model (ELM; Petty and Cacioppo 1986; Petty et al. 2005) which describes leaders’ language strategies for connecting with their audience. ELM lays the foundation for two key routes to persuasion, or how listeners process and understand the content of a message: the central and the peripheral routes (Petty and Cacioppo 1986; Petty et al. 2005). The central route is more cognitively taxing on the listener and requires significantly deeper evaluation and processing than does the more facile, peripheral route. In other words, listeners must expend more effort to parse the language and process messages when speakers use the central route, used in technical explanations requiring attention to detail. This is not the language of mass public appeal.
The peripheral route relies on emotional pitches and simplified heuristics. In general, populist leaders are likely to use the peripheral route to persuasion in their mass communication because it requires their constituents to expend minimal effort in processing the message (Fiske and Taylor 1984). Whereas in a private audience, a populist leader like Chavez might use the central route more often to persuasion among his close inner group of supporters, publicly, the peripheral route is generally more effective at persuading the masses due to the simplicity and ease in parsing the content of the message. In a sample of leaders whose countries experienced upheaval in the Arab Spring, Windsor et al. (2017), demonstrate that those who remained in office used language to encourage in-group support through positive emotions. This suggests that emotionally laden language—indicative of the peripheral route to persuasion—is a more successful strategy complex argumentation.
Consistent with what we have observed about Chavez’s language during the Aló Presidente broadcasts, he uses both positively and negatively valenced language, to varying degrees of success in bolstering his popularity. A study on political language and aggression involving cases from multiple countries representing a broad geography, Matsumoto, Hwang, and Frank (2014) find that increased usage of angry language was associated with increased use of acts of aggression against citizens. Angry language may also indicate diversionary behavior when the anger is directed at external actors such as the United States, exemplified in this 2006 United Nations speech: “The devil came here yesterday, and it smells of sulfur still today, this table that I am now standing in front of” (Chavez 2006). Recent research also supports the notion that leaders may use more hawkish behavior when trying to bolster partisan morale (Foster and Palmer 2006), which should manifest in their language through increased negative rhetoric.
Leaders likely increase their use of optimistic language to gain favor from their domestic audience. Bono and Ilies (2006) find that charismatic leaders use more positive emotion words, and that this causes their audience to also experience increased positive emotions. The audience’s elevated “mood” increased positive ratings of the leader’s performance as well as their attraction to the leader. Others have evaluated the lexico-syntactic properties of charismatic speech, including characteristics such as pronoun density and disfluencies, as well as the number of content words, that is, nouns and verbs. Bligh and Kohles (2009) argue that charismatic leadership helps to explain the mass enthusiasm felt during the 2008 election campaign, in particular the positive language used by Obama to persuade followers. Charismatic leaders can initiate social change by articulating a “powerful vision that inspires and motivates them” (Seyranian and Bligh 2008, 54). This includes the use of optimistic, positive language.
We expect Chavez would use positive language strategically to influence citizens’ perceptions of his leadership, reflecting Jagers and Walgrave’s (2007) analysis of populism in Belgian politics where they conceptualize “thin” and “anti-elitist” populism, apt definitions that describe Chavez’s rhetorical style. Armony and Armony (2005) note the Kirchner’s usage of optimistic language to maintain a mythology about a strong and forward-moving Argentina in the midst of a crippling economic crisis. They write that “Kirchner’s election expressed an apparent ‘reconciliation’ between citizens . . . and the political establishment” which helped to bolster a remarkably high approval rating of 88 percent (Armony and Armony, 2005, 48).
Expectations
From the literature on democratic accountability, leader persuasion, and discourse, we derive a set of preliminary expectations that take into consideration the competing theoretical strands in the U.S. literature of vertical accountability (Erikson, MacKuen, and Stimson 2002) and elite control of mass opinion (Zaller 1992). Because of the potential countervailing theoretical predictions and the limited literature in political science about language and persuasion, we make only general predictions about the linkages between presidential approval and leader language that expresses optimism or anger.
We expect Chavez used language that was optimistic and hopeful during periods of popular support and that this type of discourse would, in turn, help keep executive approval buoyant. From the literature of persuasion, we should expect leaders during periods of “good times” to amplify and convey the positive environment through optimistic language. With regard to the more negatively charged language of “anger,” we expect a negative relationship. It is during periods of challenge that we expect a leader, and a populist leader in particular, to turn to language that expresses anger at a foe, challenger, or “other.” Chavez often denounced domestic economic elites, political opponents, and, famously, the United States and George W. Bush. Although we expect lower approval to trigger more negative language, it is less clear what the theoretical expectation is for anger’s impact on approval. Although leaders’ optimistic language confers optimism to the general public, bolstering support via emotional contagion, it is unclear whether negative language does the opposite (Bono and Ilies 2006).
Although the exogenous economic factors are mostly included as controls, we can also make expectations regarding them that mirror those for approval and language. During “good times,” that is, high oil prices and low inflation, we expect higher approval, more optimistic language, and less anger language. As the economic environment worsens, Chavez should use more words linked to anger and fewer associated with optimism; this is particularly true with regard to oil prices as the resource provides a majority of the government’s resources.
Data and Method
A key challenge to testing the relationship between presidential approval, economic conditions, and leader discourse is a general lack of long-running series of public opinion data on executive approval outside of the United States or a few other established democracies. Although public opinion survey research has seen a dramatic expansion in Latin America over recent decades, data availability still varies widely between firms and time. Even for the available data, differences in question wording, response sets, length of time series, missing data, temporality, and sample frame have frustrated efforts to create a continues measure of approval, even for a single country. We resolve these issues by using data from the Executive Approval Project that collects and combines all available approval series for Venezuela during Chavez’s tenure in office. These raw survey marginals from various firms are “smoothed” into a quarterly time series that is comparable across time despite being derived from various data sources. 6
The approach taken by our approval data source, the Executive Approval Project, relies on a country-specific measurement model using a dyad-ratios algorithm (Stimson 1991) common in macro-opinion research in the United States (e.g., Enns and Kellstedt 2008; Erikson, MacKuen, and Stimson 2002) but which has only recently been employed in the Latin American context (Carlin, Love, and Martínez-Gallardo 2015a, 2015b). This method assumes that to the extent a given data time series is a valid indicator of presidential approval, the ratio of any two values within the series is a relative indicator of presidential approval. The algorithm uses all such dyadic ratios within a given series to estimate presidential approval values at regular (here, quarterly) time intervals. To combine N time series—ostensibly measuring presidential approval—for a given country into a single measure, each raw series undergoes this transformation, resulting in N dyads-ratio series. If these N dyads-ratio series are indeed relative indicators of presidential approval, they should co-vary where they have temporal overlap. To test whether this is true, the algorithm examines their common variance to see if the N dyads-ratio series do, in fact, tap a single latent construct—presidential approval. From this covariance, validity estimates are computed for each of the N series, and these are used to estimate the best single underlying series of latent approval. Exponential smoothing on the resulting series sharpens the estimates by removing random fluctuation due to sampling error. In the end, the process creates a quarterly time series measure of presidential approval for Venezuela. In this paper, we truncate the series to match the data availability regarding leader discourse, 1999 to 2013 (Carlin et al. 2016). 7
Specifically, following this approach produces a dataset of executive approval for Venezuela that uses three hundred unique data series gauging presidential “approval” (aprobar/desaprobar), “favorability” (favorable/desfavorable), and “ratings” (e.g., muy bien, bien, regular, mal, muy mal) of the president’s “management” (gestión), “job/work” (trabajo), “performance” (desempeño), and “image” (imagen). Excluded are questions related to vote intentions, trust or confidence, and grading systems (e.g., 1–10). If the response choice is dichotomous, the marginal of the positive response is employed. If it is trichotomous with a “regular” or neutral middle category, only the “positive” marginal are analyzed. For four-part response choices, all positive responses are summed; if there are five responses, again only positive response marginal are summed while excluding “regular” or neutral categories.
Several pieces of evidence suggest this novel approach delivers valid and reliable measures of presidential approval and thus allows us to analyze patterns of accountability and the linkage between language and approval. We find that a single dimension, theorized to be presidential approval, accounts for an average of 84.31 percent of the variance in the measurement model. Moreover, most individual input series load highly on the latent factor, over 0.90. By meeting conservative criteria for confirmatory factor analysis, these results bolster our confidence that despite differences across the input series (margins from firms), the information can nonetheless be harnessed to produce a valid measure of presidential approval in Venezuela. We are confident that distinct question wordings or response sets are not significantly affecting our results as they do not lead to systematic differences in loadings estimates.
Discourse Data
We derive linguistic data from Chavez’s television program, Aló Presidente, that include dialogue from both Chavez and journalists who ask questions. Although the majority of the transcript is Chavez’s own words, we include both sides of the dialogue, that is, both Chavez’s language and the journalists’ questions. We did this for several reasons: often the journalist will refer to a content word, that is, a proper noun or specific idea, in the question, but Chavez may not reiterate that word in his response because of repetitiveness. In addition, Chavez does the vast majority of talking during the program and in doing so, he controlled the tone, content, and direction of the discussion.
We analyzed the corpus with the Linguistic Inquiry Word Count (LIWC) program using the 2007 Spanish dictionary file (LIWC, Pennebaker, Booth, and Francis 2007), which classifies the words into dozens of linguistic and psychological categories tapping social, cognitive, and affective processes. The 2007 Spanish LIWC dictionary contains over four thousand words and is a translation of the English LIWC 2007 dictionary. LIWC does not rely on syntax or sentence structure; rather, it uses a “bag of words” approach, returning values for the proportion of words per category from the source text.
From this process, the LIWC software employing the Spanish dictionary generates the percentage of Chavez’s broadcast for a quarter that used language involving words judged, via the dictionary, to convey optimistic affect (optimism). Likewise, we used the process to calculate the percentage of language from broadcasts within a quarter that used words that have an anger affect (anger). Our linguistic measures are the percentage of Aló Presidente broadcast transcripts for a given quarter that use language conveying optimism or anger.
Economic Factors
We include two measures of economic activity that are likely to affect executive approval and the type of language used by a Venezuelan leader. No other country in Latin America is as economically reliant on the export of a single commodity as Venezuela, the tenth largest oil exporter in the world. Oil constitutes 95 percent of the country’s exports and accounts for around 15 percent of gross domestic product (GDP; Organization of the Petroleum Exporting Countries 2016). Due to the nationalized oil industry, petroleum exports comprise the bulk of government revenues, accounting for upward of 70 percent. With social services and wealth redistribution a key source of government legitimacy, oil revenues played a pivotal role in Chavez’s popularity.
Other economic factors may also influence both discourse and approval. Work by Carlin, Love, and Martínez-Gallardo (2015) show inflation plays a central role in shaping public views of presidents in Latin America. During periods of higher inflation, citizens hold executives accountable for failing economic policies while rewarding presidents with support during good economic times. Recent literature demonstrates the key role economic performance plays in shaping public support of executives (Carlin, Love, and Martínez-Gallardo 2015a; Johnson and Schwindt-Bayer 2009; Pérez-Liñán 2007; Singer 2011a, 2011b). It is also likely leaders will shift the nature and subject of their speech during economic downturns. As Powell and Whitten (1993) and Carlin, Love, and Martínez-Gallardo (2015b) show, leaders facing divided government can effectively shift blame for economic failures onto the legislature. Part of this blame shifting behavior should occur through discourse, as is seen following exogenous events like natural disasters (Windsor, Dowell, and Graesser 2014). To control for these potential factors, we include the percent quarterly change in the consumer price index for Venezuela (World Bank’s World Development Indicators). As a robustness check, we estimated the models below with additional controls for unemployment, growth, and national elections. The results are substantively similar, but we exclude them because the relatively short available time series limits the degrees-of-freedom for model estimation. Descriptive statistics are reported in the online appendix.
Method
We use a series of vector autoregression models (VAR). Although conceptually approval and leader discourse may be endogenously related to each, we face two challenges in using a VAR approach to assess these potential relationships: first, we have a relatively short series. In the VAR setup, each endogenous variable requires a full model with multiple lags. To address the potential degrees-of-freedom issue, we model the relationship between approval and optimism separately from approval and anger. The exogenous controls are limited to two key economic variables, global oil price and inflation. Finally, we selected the lag-order structure of the models based on lag-order tests using likelihood ratio, AIC, final prediction error, and HICQ. The results strongly indicate that a relatively short two-quarter lag is the appropriate specification.
The second challenge we face is gaps in the data. Due to corrupted or missing Aló transcripts, we have four, non-sequential quarters with no data. In a cross-sectional setup, this would likely create little bias, but because of the autoregressive structure of time series analysis, such holes are an impediment. To create an uninterrupted series, we employ Hermite cubic interpolation to fill the missing quarters. 8 The descriptive graphs of the time series below are based on the interpolated series for Optimism and Anger.
Finally, Augmented Dickey-Fuller unit-root tests indicate that both of the language markers, Anger and Optimism, are stationary while the test for Approval shows it is possibly a non-stationary process. Because of the potential for spurious correlation in a VAR from non-stationary series, we run alternative specifications using first differences of the variables, making them stationary, in a VAR structure for non-cointegrated series (Approval, Anger) and a vector error-correction model (VECM) for cointegrated series (Approval, Optimism). Lag structures were selected using model fit statistical tests. Both approaches give results substantially identical to those reported in the paper giving us confidence that the VAR results reported here are not a result of issues of non-stationarity (results for the VECM and first-differenced VAR are reported in Online Appendix Tables A2 and A3, and Figure A3).
Results
Figure 1 displays the trends of Chavez’s approval and his use of optimistic language during Aló Presidente broadcasts. Although some large variance in the discourse measure appears to be random, there appears to be a relationship between the two series. Whether changes in approval precede changes in optimistic language or the reverse is unclear. Figure 2 displays a bit more ambiguous relationship between approval and use of “anger” language. We see a fair amount of random noise in the discourse series and a more limited relationship between the two series. Both figures provide a hint that there is a relationship between the language used by Chavez in his public appearances on Aló and his popular approval. To assess whether the series are statistically correlated, while accounting for economic conditions in Venezuela, we turn to the VAR models in Tables 1 and 2. 9

Approval and “optimism” during Chavez’s tenure.

Approval and “anger” during Chavez’s tenure.
Vector Autoregression—Endogenous Variables: Approval, Optimism (N = 50).
Model 1: R2 = .88; Model 2: R2 = .41
p < .1. **p < .05. ***p < .01 (two-tailed test).
Vector Autoregression—Endogenous Variables: Approval, Anger (N = 50).
Model 1: R2 = .89; Model 2: R2 = .23
p < .1. **p < .05. ***p < .01 (two-tailed test).
Table 1 shows Chavez’s executive approval is predicted by his use of optimistic language; however, we find no evidence that approval shapes his use of language. We also find oil prices and inflation affect approval but have no influence on language. When Chavez uses more optimistic language, we see higher levels of approval in the following quarters. A Granger causality test for optimism causing approval has a p value of .03 while we find no evidence of approval Granger causing the use of optimistic language.
For anger, Table 2 shows a negative relationship between approval and “anger” language. Like with optimism, anger language predicts approval but is unaffected by approval. Again, exogenous economic factors strongly predict approval with only a minor effect on language. A Granger causality test for anger also shows that anger language Granger causes presidential approval for Chavez but approval does not predict the use of language with an anger affect.
Finally, Figures 3 and 4 show orthogonalized impulse-response function (IRF) graphs for optimism and anger, respectively. Orthogonalized IRFs, which are a subset of structural IRFs, show how an unexpected positive increase of one standard deviation, the impulse, affects another endogenous variable over a period of time (ten quarters in this case) through the model structure. 10 As VAR models are inherently dynamic, impulses do not have immediate, limited, and linear effects. Both figures below show significant effects on approval from increases in the language markers that are not immediate and last approximately a year.

Orthogonalized IRF—Impulse: Optimism; Response: Approval.

Orthogonalized IRF—Impulse: Anger; Response: Approval.
Although we are able to use interpolation to have a fifty-two-point time series, the VAR models may be somewhat underpowered with the necessary inclusion of economic controls. With little evidence that approval shapes language, we turn to a Koyck specification to estimate more parsimonious, yet exogenous, models in Table 3. We employ a Koyck model structure after estimating the more general autoregressive distributed lag (ADL) model and finding no significant direct short-run effect of the lagged exogenous variables on executive approval (De Boef and Keele 2008; Williams and Whitten 2012). The model fit of the more general ADL model was nearly identical to the more restricted model. The restricted model also helps ensure that the results from the VAR (or a full ADL) are not the results of an over-fitted model that chases random variation, or noise, in the series rather than structural or causal patterns. 11
Koyck Model of Approval and Optimism (N = 51).
p < .1. **p < .05. ***p < .01 (two-tailed test).
As the effects of a change in the independent variables in a particular quarter reverberate through the approval time series via the autoregressive term, the full effect of the variables are not immediately observed. The long-run multiplier depicts the full effect of a change in an exogenous variable through all subsequent quarters in the series (De Boef and Keele 2008; Williams and Whitten 2012). Thus, in Table 4, we also estimate the long-run multipliers for the variables in the models estimated via the Bewley (1979) transformation. As was seen in the VAR models, both the language and economic variables are consistent long-run predictors of approval.
Koyck Model of Approval and Anger (N = 51).
p < .1. **p < .05. ***p < .01 (two-tailed test).
The evidence from both the VAR and Koyck models indicate leader discourse can shape public approval, supporting the contention that elite control of messaging in a populist regime may strongly limit vertical accountability. The Venezuelan public’s view of Chavez was very much a product of his carefully crafted messaging. Of the two language markers, it appears from both the VAR and Koyck models that Anger has a greater influence. In both specifications, anger-toned language has a larger effect on approval than optimistic discourse. 12 Likewise, the results show Chavez’s language was fairly immune to the public’s view of his job performance. This finding is also the basis for a key caveat for interpreting the model results: Chavez’s language may be a leading indicator of future political challenges. That is, Chavez may be shaping his language less in response to his approval but in response to anticipated political challenges on the horizon.
Conclusion
In this paper, we use new presidential approval and discourse data to describe how one of the world’s leading populists, Hugo Chavez, managed political affairs during his tenure in office, including during times of political and economic crisis. Our results support the claim that he employed classic populist techniques, in particular increased optimistic language, to manage public approval. Likewise, when he turned to anger language of the peripheral route, we also saw shifts in approval due to differing levels of anger language use. Aló Presidente testifies to Chavez’s own belief in his ability to build and manage support through language and discourse.
The overall results indicate that an archetypal populist leader such as Chavez has immense power to shape public support (for good or ill) through discourse via media. This would indicate Zaller’s RAS model may have much to tell us about populist leaders and that the careful systemic feedback loops that help stabilize vertical accountability in the United States are irrelevant in a country like Venezuela. The role of public “mood” or latent public opinion appears to have little impact on how Chavez used his gifts of populist discourse. However, we also temper this claim. The consistent significant importance of exogenous economic factors on Chavez’s language points to the possibility that language is a leading indicator. A leader may leverage populist rhetoric in the face of changing exogenous factors that have not yet affected the leader’s approval and is not captured by the standard measures of economic performance included in the models. Both explanations are not mutually exclusive; leader language may both shape approval and be a leading indicator of it. As Zaller (1992) argues, leaders should anticipate changes in public opinion. Leaders attempt to foresee the (political) future, often with much error. Overall, the findings of this paper highlight the importance of the structure of discourse for how a populist leader manages public support and maintains power in a semi-democratic regime. Less clear is whether a populist leader in a more established democracy can wield the same tools of populist rhetoric as effectively as Chavez. The recent rise of populism in the United States (Trump) may soon provide a potentially illuminating case to examine the efficacy of populist discourse for mass control in a more open and pluralist environment than the case of Venezuela.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship and/or publication of this article: Minerva Initiative, Department of Defense (PI: Leah Windsor) FA9550-14-1-0308.
Supplemental Materials
Supplemental materials for this article are available with the manuscript on the Political Research Quarterly (PRQ) website.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
