2. Literature Review and Latin American Trends

Even if invalid voting has a long history—having de facto existed since the adoption of the secret ballot in national elections—it receives significantly less attention from electoral scholars than related behaviors such as voter abstention. One reason might lie in some confusion between the two, and another is the inherent ambiguity of invalid voting as a political act that somehow constitutes both voting and its opposite. Most work on invalid voting focuses on institutional or individual-level causes, often interpreting this phenomenon as unhealthy civic behavior to be corrected. This contrasts with evidence from Latin America, where invalid voting is widespread and routinely used as a gesture of protest against institutionalized politics.

Conceptually, academic work differentiates between unintentional invalid voting—mistakes made in the ballot cabin—and intentional invalid voting—cast with some kind of protest motivation. The former typology is known as the “voter error” hypothesis and usually works through two different mechanisms. One is different ballot designs, assuming that more complexity results in more voting mistakes (Herron and Sekhon Reference Herron and Sekhon2005; Herrnson et al. Reference Herrnson, Hanmer and Niemi2012; Pachón et al. Reference Pachón, Carroll and Barragán2017). This assumption implies that technological advancements should reduce invalid voting (Desai and Lee Reference Desai and Lee2021) and is accompanied by the observation that invalid voting is generally segmented by voters’ education level (McAllister and Makkai Reference McAllister and Makkai1993; Fatke and Heinsohn Reference Fatke and Heinsohn2017). Yet, this relationship is nondeterministic, as the introduction of simplified ballots in 1990s Brazil, for example, did not reduce invalid ballots. Moreover, invalid voting can still grow, even after literacy has spread globally. An alternative mechanism for involuntary invalid voting connects the large choice sets associated with party system fragmentation with the impossibility of making a choice (Moral Reference Moral2016; Cohen Reference Cohen2018a; Lysek et al. Reference Lysek, Lebeda and Kouba2020).

On the other hand, intentional blank and spoiled ballots are usually attributed to alienated, disenchanted, disempowered, or unsupportive voters (Cohen Reference Cohen2018b). Across many studies of invalid voting, blank and spoiled ballots are considered an act of political protest (Power and Roberts Reference Power and Timmons Roberts1995; Zulfikarpasic Reference Zulfikarpasic2001; Power and Garand Reference Power and Garand2007; Uggla Reference Uggla2008; Cohen Reference Cohen2018b; Hennau and Ackaert Reference Hennau and Ackaert2019), which has also been observed in less democratic or even fully authoritarian regimes along a spectrum that goes from Serbia all the way to Azerbaijan and Iran (Samii Reference Samii2004; Herron Reference Herron2011; Obradović-Wochnik and Wochnik Reference Obradović-Wochnik and Wochnik2014). To organize the existing body of work on the subject, a recent publication has tested the validity of explanatory factors through mutually exclusive categories (Kouba and Lysek Reference Kouba and Lysek2019). This meta-analysis found that a majority of studies empirically confirm only three hypotheses: compulsory voting, party-system fragmentation, and low democratic quality are all associated with higher invalid voting. Evidence from other variables relative to the economy, political competition, and national institutions is contradictory at best, which makes the investigation of new explanatory paths necessary.

In a Latin American context, the invalid voting scholarship saw early progress through notable studies that combined aggregate electoral data from the 1980–2000 period with institutional, socioeconomic, and protest-related variables, showing the importance of factors such as urbanization and income inequality for invalid voting levels (Fornós et al. Reference Fornós, Power and Garand2004; Power and Garand Reference Power and Garand2007). Recently, it has been also found that Latin American voters casting an invalid ballot are alienated from politics, but do not have antidemocratic tendencies (Cohen Reference Cohen2018b). In parallel, the same author captured the influence of political fragmentation through a dynamic model, finding that larger candidate sets are connected to more invalid voting, but that new actors reduce spoiled and blank ballots (Cohen Reference Cohen2018a). To this effect, this study explores blank and spoiled ballots cumulatively, because there is much confusion in the way that these two behaviors are calculated on an aggregate level in different countries and we lack empirical evidence that leads us to consider them as different. This choice is justified by the ample evidence that these two political behaviors are extremely similar and generally vary together in quantitative studies (Aldashev and Mastrobuoni Reference Aldashev and Mastrobuoni2019; Lioy Reference Lioy2021b), and that both are alternatively used with explicit protest intent in Latin America (Cisneros Yescas Reference Yescas and Isaac2013; Sonnleitner Reference Sonnleitner2019; Palacio Vélez Reference Vélez and Camilo2022).

If regional-level studies on the subject are relatively scarce for Latin America, in several national contexts invalid voting has been explored as an alternative to partisan voting. Given the impossibility of covering the whole region, this paragraph covers some notable cases. For instance, the recent emergence of invalid voting in Mexico following the 2006 elections has sparked a lively debate over its general validity, impact, and different forms (Cisneros Yescas Reference Yescas and Isaac2013; Freidenberg and Aparicio Reference Freidenberg and Javier Aparicio2016; Sonnleitner Reference Sonnleitner2019). Similarly, in Ecuador, invalid voting has been associated with the failure of traditional parties to elicit electoral support (Machado Puertas Reference Puertas and Carlos2007) and recently with the grievances of the non-Correista left (Sandoval Reference Sandoval2021). The sheer magnitude of invalid voting in Brazil has also been inescapable for the local scholarship, which has to dispel the false conviction that invalid votes will lead to a repeat election (Soares and D’Araujo Reference Soares and Celina D’Araujo1993; Borba Reference Borba2008; Chiarello and Sarai Reference Chiarello and Sarai2017; Cavalcante Nepomuceno and Seixas Costa Reference Nepomuceno and Thyago2019). Finally, in Colombia, blank ballots’ political relevance has been explicitly recognized by the local academia as the expression of popular discontent and a potential driver of change (García Sánchez and Cantor Reference García Sánchez and Cantor2018; Soracá Becerra Reference Soracá Becerra2018; Otálora Heredia and Díaz Martínez Reference Heredia, Paula and José Díaz Martínez2019; Palacio Vélez Reference Vélez and Camilo2022).

A historical overview of invalid voting in Latin America, examining its magnitude and salient variations, is then performed through electoral data from the Institute for Democracy and Electoral Assistance (IDEA) elections dataset, cross-referenced with Nohlen (Reference Nohlen2005). Figure 1 shows the evolution of invalid voting across the continent.Footnote ¹ The trend line traces the rise of invalid voting across Latin America from a 5% average in 1950 to 10% in 1999, followed by a slight decline and recovery by 2020. This volatility vis-à-vis the stability of national institutional arrangements highlights the inherent political component in invalid voting: different ballot designs or electoral systems may explain a certain level of invalid voting, but changes in invalid voting are the consequence of shifting political conditions. A scholarly conversation about these high invalid voting levels avoids dangerous neglect, as when too many voters decide not to choose a political party, democratic representation deteriorates. One explanation for these long-term regional trends is that the number of Latin American democracies increased as the Cold War ended, but widespread frustration with the new representative institutions also rapidly appeared (Benton Reference Benton2005). Even if democracy alone was soon deemed insufficient for creating equality or development, street protests fell across the region (Schatzman Reference Schatzman2005).

Figure 1. Invalid Voting Trends in Latin America

Note: Yearly averages, national-level legislative elections.

Table 1 reports general invalid voting statistics for legislative elections that took place throughout the 18 countries in the dataset. The left side of the table shows that the electorates of Brazil, Ecuador, Guatemala, Peru, and post-1991 Colombia cast invalid votes at much higher levels than the rest of the continent. All these party systems are highly fractionalized, with many parties gaining access to parliament, which indicates that high levels of invalid voting in Latin America are connected with political fragmentation and partisan volatility. Additionally, a sample-wide average of 9.1% for invalid voting shows its presence even where political stability is taken for granted, such as in Chile and Costa Rica.

Table 1. Invalid Voting in Latin American Legislative Elections

Note: Countries with average invalid voting levels > 10% are in bold. *After the 1991 Constitution, the Colombian average is 14.34%.

Note: Each country’s largest variation is marked in bold.

Since extreme cases provide additional insight beyond average levels, the right side of table 1 summarizes the 18 largest positive variations of invalid voting. As expected given their average levels, Brazil, Ecuador, Colombia, and Peru are all featured, but the presence of eight different countries confirms that sudden surges of invalid ballots can occur in stable party systems, as happened in 1997 in Chile, where the debate on invalid voting has been recently revived (Ponce de León Solís Reference Ponce de León Solís2019). In Central America, this phenomenon has been prevalent in the hyperfragmented Guatemalan context, where antipartisan politics are strong (Azpuru and Blanco Reference Azpuru and Blanco2008). Even if invalid ballot levels and volatility increased over time, sudden surges happened at political junctures, for example predating 1960s authoritarianization. Notably, in Argentina, despite low average invalid voting levels (6.5%), there have been three large variations, two of which took place when Peronism was banned (Baily Reference Baily1965; Snow Reference Snow1965) and a third during the economic crisis of 2001 (Escolar et al. Reference Escolar, Calvo, Calcagno and Minvielle2002; Parma Reference Parma2013). Figure 2 shows trends in six notable Latin American countries.

Figure 2. Invalid Voting Trends in Six Latin American Countries

3. Conceptualization, Research Questions, and Hypotheses

Generally, research on the topic still suffers from some persisting confusion about what the act of casting invalid ballots represents for voters. Even if elections are considered as the core of representative democracy, they are not the only relevant form of political participation. Unfortunately, the study of alternative forms of voting has long remained marginal because of a subfield separation between scholars of institutional politics and those of social movements, which illustrious scholars have lamented before (McAdam and Tarrow Reference McAdam and Tarrow2010). Conceptually, invalid voting sits between two modes of political participation (labeled I and II in table 2) that can be called conventional and unconventional, or institutional and noninstitutional (van Deth Reference van Deth2014). Invalid voting falls into a gray area because it has the formal components of electoral behavior that associate it with partisan voting, and some informal ones—such as writing on the ballot the name of a celebrated national hero—that bear resemblance to protesting in the streets. These issues in the conceptualization of invalid voting are not trivial, because different conceptualizations—institutional versus informal—result in divergent scholarly approaches, which are summarized as institutional, social, or protest-related explanations by McAllister and Makkai (Reference McAllister and Makkai1993), whose work draws heavily from the voter turnout literature.

Table 2. Concepts, Types, and Typical Modes of Political Participation

Source: van Deth (Reference van Deth2014, 361).

On the institutional side, invalid voting is influenced by the structure of the political offer, visible through the relationships of strength between political parties, which is notoriously crucial to all voter choices (Wessels and Schmitt Reference Wessels and Schmitt2008). Rather than being disgruntled against institutions or opposed to representative democracy (Cohen Reference Cohen2018b), protest voters might simply be dissatisfied with a specific election’s political equilibria. In elections where one’s preferred party is weak and unlikely to win, invalid voting is a natural alternative, together with voting for marginal parties, which is positively correlated with spoiled and blank ballots (Uggla Reference Uggla2008). At their extreme, voters cast invalid ballots as an “impossible vote” if their desired option is outlawed, as happened with Peronismo in 1960s Argentina (Baily Reference Baily1965).

Unlike abstention—a passive behavior even when intentional—invalid voting implies the act of casting a ballot. The average invalid voter is politically active, but often does not feel represented in the current climate, as opposed to abstainers who are generally less involved in politics (García Sánchez and Cantor Reference García Sánchez and Cantor2018). As noted by Martinez i Coma and Werner (Reference Martinez i Coma and Werner2019), abstention is an “exit” behavior, a way of rejecting the system, while invalid voting still represents a “voice” option, a form of political expression. Note how a notorious taxonomy of protest voting includes insurgent party, tactical, BNS (blank, null, spoiled), organized, and officially sanctioned protest voting (Alvarez et al. Reference Alvarez, Roderick Kiewiet and Núñez2018), where these categories can also overlap.

From these considerations, a question arises: if invalid ballots are part of a “repertoire of contention,” do they also act as a complement to public demonstrations and antigovernment mobilization? If blank and spoiled ballots are protest votes, then one can hypothesize a positive relationship with the occurrence of actual political protests. One problem for testing this assumption lies in empirically connecting these behaviors, with quantitative studies using proxies such as economic crisis, unemployment, hyperinflation, a lack of political legitimacy, and democratic deterioration.

In the academic literature, an authoritarian transformation of national politics is thought to be a significant driver of invalid voting, and previous studies found a robust negative relationship between democracy levels and blank/spoiled ballots (Power and Garand Reference Power and Garand2007; Cohen Reference Cohen2018a; Singh Reference Singh2019; Carlin and Moseley Reference Carlin and Moseley2022). If a desire to defend democracy motivates voters to cast invalid ballots in protest, then changes should compensate for variations in democracy levels. In Latin America, 1990s Peru offered the clearest example of this dynamic, as the rise of Fujimorismo was accompanied by a massive surge in blank ballots (Durand Reference Durand1996; Carrión Reference Carrión2001; Franco Reference Franco2001; Zolezzi Chocano Reference Zolezzi Chocano2006).

Recently, corruption has also been used as a proxy for protest in voting models, because a belief that politics is rigged can easily trigger antagonistic political behavior such as invalid voting. Generally, high perceptions of political corruption among government officials can hinder the efforts of political elites to bring people to the polls if they are perceived as insincere (Fossati and Martinez i Coma Reference Fossati2020). In terms of a direct relationship, the literature has found that at the aggregate level higher perceived corruption levels increase invalid ballots under compulsory voting (Martinez i Coma and Werner Reference Martinez i Coma and Werner2019). Cohen (Reference Cohen2018b) has also used corruption perceptions as a control variable for testing whether antisystem motivations for invalid voting were in reality just a sophisticated reaction to perceived corruption. Another recent regression-based study of invalid voting chose not to include corruption (Lysek et al. Reference Lysek, Lebeda and Kouba2020) and justified this choice by claiming that corruption scores are too correlated with per capita GDP and Freedom House values.

In addition to democratic deterioration and political corruption, economic conditions, inflation, and unemployment rates are frequently featured in quantitative modeling as sources of protest behavior that lead disgruntled economic voters to invalidate their ballots (Power and Roberts Reference Power and Timmons Roberts1995; Driscoll and Nelson Reference Driscoll and Nelson2014; Dejaeghere and Vanhoutte Reference Dejaeghere and Vanhoutte2016; Aldashev and Mastrobuoni Reference Aldashev and Mastrobuoni2019; Adeleke et al. Reference Adeleke, Alabede and Amusan2022).

Even if the use of these proxy variables is common, the academic literature includes some indication of a direct link between protest participation and widespread invalid voting, which can be reasonably traced both for antigovernment protests and for labor strikes, which may be more or less common in different contexts. In Latin America, country-level evidence supports this connection, which is yet to be tested at the regional level. For example, in the 1980s the Ecuadorian and Brazilian social scientists commenting on invalid voting growth also discussed the simultaneous emergence of intense public protests (Pérez Sáinz Reference Pérez Sáinz1987; Moisés Reference Moisés1990; Soares and D’Araujo Reference Soares and Celina D’Araujo1993; Sánchez Parga Reference Sánchez Parga1994). Both behaviors—invalid voting and protest participation—were seen as a response to the insufficient democratization of Ecuador and Brazil.Footnote ² The Brazilian case also helps to establish this connection in recent years, given how a new rise in invalid voting coincided with the massive protests that led to Dilma Rousseff’s impeachment (Cavalcante Nepomuceno and Seixas Costa Reference Nepomuceno and Thyago2019).

From this discussion, a first research question can be drawn and made into specific hypotheses connecting types of protest behavior and its root causes. In other words, the hypotheses test whether invalid voting dynamics are connected to those of actual antigovernment protests and waves of labor strikes, or whether they are a reaction to changes in national levels of democracy and corruption.

Beyond political protest and its causes, one can explore the institutional component of the story, for which the literature proposes an association between electoral competition and invalid voting (McAllister and Makkai Reference McAllister and Makkai1993). The main hypothesis is that invalid voting is a protest behavior that responds to political fragmentation or to waning political competition. Caution is needed, because the empirical relationship between institutional political behaviors and invalid voting is undetermined (Kouba and Lysek Reference Kouba and Lysek2019) and may operate in the opposite direction (Aldashev and Mastrobuoni Reference Aldashev and Mastrobuoni2019).

First, the most common proxy for party competition is the margin of victory between the two largest parties: in more competitive electoral contexts votes are believed to be more valuable, not to be wasted by invalidating them. Second, lack of political alternation keeping the same party in power might correlate with growth in invalid ballots, because it reduces national-level political competition by entrenching a government clique. This goes along with previous studies that have shown that invalid voting is a last resort when no exit or voice option is viable (Hirschman Reference Hirschman1970; Hooghe et al. Reference Hooghe, Marien and Pauwels2011), or a protest tool against clientelist, antidemocratic actors (Carlin and Moseley Reference Carlin and Moseley2022). Third, relevant to political competition and related to invalid voting is casting a ballot for political outsiders, which is another way to protest institutional politics. Following the results of Uggla (Reference Uggla2008), one can expect invalid voting to rise and fall together with votes for marginal parties. A recent study confirmed this association, by showing how invalid voting predicts electoral support for Italian populist parties (Aron and Superti Reference Aron and Superti2022). Notice how small-party strength also captures party system fragmentation (Fatke and Heinsohn Reference Fatke and Heinsohn2017), and the more fragmented the party system, the harder the choice process, ultimately resulting in more invalid ballots (Cunow et al. Reference Cunow, Desposato, Janusz and Sells2021).

4. Research Methodology and Variable Operationalization

The decision to assemble a novel dataset of Latin American elections up to 2021 is justified by the lack of up-to-date empirical results for the whole region. A Latin American focus is also fit for theory testing because of the variation provided by (1) enforced compulsory voting in half of the countries, (2) patterns of protest and worker unionization, and (3) different levels of democratic consolidation and corruption. Furthermore, invalid voting is common across the region, where it often channels popular dissatisfaction toward the partisan alternatives on the ballot, and where the amount of blank and spoilt ballots cast often shows intense volatility between elections.

A good empirical model of invalid voting should investigate the overarching associations between formal and informal political behavior, but it is often hard to compare data from different countries. This paper addresses these issues through a dynamic model: the dependent variable is the change in the percentage of invalid ballots between consecutive legislative elections and the independent variables are also in the differences. This choice enables the researcher to test the factors altering voters’ decision to invalidate their ballot, and previous electoral studies have adopted a dynamic approach for congressional turnout in the United States (Clouse Reference Clouse2011) and for invalid voting in Latin America (Cohen Reference Cohen2018a). Using variables in differences presents the following advantages over a static specification:

1. 1. It sets aside each country’s baseline level of invalid voting, avoiding reliance on institutional variables or fixed effects specifications.

2. 2. It makes different countries’ political contexts more comparable, as it looks at changes in citizens’ behavior that depend on a changing national environment.

3. 3. One can hardly argue that changes in invalid ballots are involuntary or simple errors. When many suddenly cast invalid votes, they “preferred not to” vote for existing parties, resorting to passive resistance.Footnote ³

4. 4. Looking only at changes blurs the lines between blank and spoiled ballots, allowing their cumulative use as two different manifestations of invalid voting.

Even if a fixed effects model seemed unnecessary, the Hausman specification test confirmed that a standard panel model with random effects is to be preferred for the regression analysis, with a p-value > 90% across different model specifications. The model is also run with robust standard errors, to make up heteroskedastic error terms across the sample. Finally, the dataset was checked for accuracy and consistency across sources to ensure homogeneity.

To overcome the existing literature’s limits and fragmentation, this paper tests for the existence of a direct link between simultaneous trends in invalid voting and mass protest. A novel events-based dataset including yearly assessments of the intensity of antigovernment protests and workers’ strikes throughout all of Latin America over the past 40 years is what made this study possible (Martínez Reference Martínez2021). In practical terms, this new dataset quantified protest occurrence and intensity in a given year on a uniform basis, using the Latin American Weekly Report as a cross-national source.

Here the main independent variable is measured as the change in the intensity of public protests (H1) between the two electoral years under scrutiny, normalized on a 0 to 4 scale. After coding the static variable for both in a given election year, the variable in differences can then range from −4 to +4. A positive variation indicates that the intensity of protests in the electoral year increased in comparison with the previous election, and vice versa for a negative variation. In the regression model, the change-in-protest variable is then made to interact with variables that measure changes in democracy levels, power alternation, political corruption, and compulsory voting. This captures whether different behaviors influence each other and whether public protests and invalid voting correlate only under some conditions. For example, it would be important to know if the emergence of massive demonstrations results in more protest voting only when political corruption spirals out of control or when an authoritarian transition is feared. The same goes for compulsory voting—thought to increase blank and spoiled ballots by raising the cost of abstention—which could perhaps reinforce the relationship between invalid voting and protest participation.

As mentioned in the theoretical section, connected with, and alternative to, antigovernment protests is the emergence of workers’ strikes (H2)—also included in the same dataset (Martínez Reference Martínez2021)—which enters the model as a variable capturing variations on a −4 to +4 scale. The big difference between the two variables lies in the strong regional disparities in unionization levels, which should make workers’ strikes a less reliable factor. To address a possible criticism, and to be jointly effective, the indexes for protests and strikes should have limited overlap. A correlation test for the two variables in differences returns a very low value of ρ = 0.0892, which justifies the inclusion of this additional variable.

Coming now to H3, the model uses the difference in the Freedom House indexFootnote ⁴ to capture whether variation in democracy levels has an independent effect on invalid voting between consecutive elections as an independent variable. Knowing that this index grows when a country becomes more authoritarian and falls when it democratizes, note that fewer than half of observations are “0” (stability), as democracy in Latin America changes quite frequently. Also notice how dramatic changes in the number of invalid ballots have already occurred in highly democratic contexts (Lioy Reference Lioy2021b).

As for H4, any serious study of protest and voting must include a variable capturing corruption perceptions, taken here in changes between electoral years in the Political Corruption Index included in the V-Dem dataset (McMann et al. Reference McMann, Daniel Pemstein, Teorell and Lindberg2016). Operationally, this index summarizes six different types of corruption and ranges from 0 to 1, where a higher number indicates more corruption. This is a common proxy adopted as a driver of invalid voting in the absence of a direct assessment of collective protest behaviors, and here its inclusion tests for whether corruption has a separate impact on invalid voting even once protests and strikes are taken into account. This is because those who protest in the street could not be the same people who cast invalid ballots, with both behaviors triggered instead by heightened political corruption levels in the country.

To answer its second set of hypotheses related to electoral competition, this work includes the change in the victory margin between consecutive legislative elections (H4), assuming a growth of invalid voting when the gap between parties expands, and vice versa for closer elections. Second, another competition-related variable captures power alternation in the party of government (H5). It is a simple dichotomous variable taking a “1” value for a change in power and a “0” for an incumbent victory, assuming that voting for the institutionalized opposition substitutes more informal political behaviors like invalid voting. Finally, the outsiders’ component is measured as the variation in the vote shares of all parties outside the top three. The expectation is for a positive relationship with invalid voting (H6), given how small-party voting constitutes a form of protest voting against the largest parties.

Coming now to the control variables, within a dynamic model, institutional transformations should still matter. For example, compulsory voting is universally considered to increase invalid voting levels through higher abstention costs (Reynolds and Steenbergen Reference Reynolds and Steenbergen2006; Power and Garand Reference Power and Garand2007; Uggla Reference Uggla2008; Cohen Reference Cohen2018a; Singh Reference Singh2019). Here, changing compulsory voting rules are expressed as a [−1, 0, +1] trinomial variable taking a positive or negative value when they are adopted or dropped (most observations will therefore be a 0). It is also established that in the election after the “founding” or “post-transition” election after an authoritarian period, voter turnout levels generally fall, as the initial mobilizing enthusiasm wanes (Kostelka Reference Kostelka2017). The model tests for its possible negative effect on invalid voting, given how invalid voting increases when turnout is near universal (Kouba and Lysek Reference Kouba and Lysek2016). The same goes for including the change in concurrent elections, which should see a positive gap in invalid voting between general and legislative-only elections, in countries holding both.

Moreover, the scholarship has shown that whenever the electoral registry adds new potential voters, it includes many electorally inexperienced citizens, less likely to go to ballot stations (Franklin Reference Franklin2004). This negative relationship for aggregate voter turnout was confirmed in a Latin American context by Kellam (Reference Kellam2013), which makes its absence from most invalid voting models rather puzzling.Footnote ⁵ Among them, only Power and Roberts (Reference Power and Timmons Roberts1995) use the percentage of enfranchised voters as an independent variable. The literature also considers literacy levels important, as less literate voters might be less able to choose in the ballot cabin (Zulfikarpasic Reference Zulfikarpasic2001), even if empirical evidence is mixed (Kouba and Lysek Reference Kouba and Lysek2019). Yet, in a dynamic model, changes in literacy are an ineffective proxy, because they show a positive trend. This work fills the gap, assuming a positive relationship between trends in registered voters and changes in invalid voting.

Finally, existing studies use the change in the growth of gross national product between consecutive elections and the change in inflation between consecutive elections as economic correlates of protest and invalid voting. This work follows their lead, and their simultaneous presence alongside the protest variable checks whether its correlation with invalid voting works even at parity of economic conditions. Conversely, including a change-in-unemployment variable would have sacrificed too many observations, given the spotty availability of labor statistics across Latin America.

To ensure coherence across theoretical, methodological, and empirical components, this paper focuses on legislative elections, where political parties have more permanence than presidential candidates. Although Latin American political parties have a comparatively short life span (Roberts Reference Roberts2013), patterns of partisan competition survive in consecutive elections and often longer. Table 3 offers summary statistics.

Table 3. Descriptive Statistics

Regarding sources, all electoral data comes from the IDEA election dataset and is cross-referenced with Nohlen (Reference Nohlen2005), which supplies data for partisan shares until 2003. To solve discrepancies, and for post-2003 party/coalition shares, national electoral tribunals’ websites and official reports were used. Economic variables come from the Penn World Tables, with missing data from the central banks of Argentina, Brazil, Nicaragua, and Venezuela. The corruption data is taken from the V-Dem dataset. The initial sample included 220 election-on-election variations of invalid voting, with 36 observations lost through founding elections and authoritarian takeovers. Including public protests and workers’ strikes further reduces the number to 148, as their source dataset (Martínez Reference Martínez2021) begins in the late 1970s.

Before proceeding with the regression analysis, a series of tests were run for the possible presence of endogeneity in the data, because electoral outcomes of different kinds happen simultaneously, as the combined effect of voters’ individual decisions. The tests of endogeneity compared two similar models: one with changes in invalid voting as the dependent variable and one where one of its determinants was used as the dependent variable, with invalid voting featured as an explanatory factor.

The results of the tests showed that two of the main variables—changes in the victory margin and changes in outsider party vote—are endogenous to the outcome of interest. These two variables were accordingly removed from the general models to avoid distorting the estimators, meaning that it will not be possible to test H4 and H6 through the regression model and the impact of political competition will be measured through H5. Crucially, because voting for outsider parties shares similar determinants to invalid voting, the two are strongly related. Conversely, the tests that used power alternation and changes in the intensity of public protests or workers’ strikes as dependent variables returned a p-value of zero, meaning that these factors are unrelated to the error term.

5. Empirical Results

Table 4 summarizes the results of the regression analyses, with models including the same core combination of variables encompassing transformations in public protests, their causes, electoral competition, and national institutions featured in column A. Specifications in models B–E also incorporate interaction terms with public protest variations.

Table 4. Determinants of Invalid Voting Variations in Latin America

Note: Random effects GLS regression, panel-robust standard errors in parentheses. °p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001.

First of all, empirical evidence from the regression models clearly supports H1, finding that changes in the intensity of antigovernment protests directly affect invalid voting variations in the direction predicted. The effect size is already important in model A, where a change from the baseline (0) to intense public protests (4) between consecutive elections is linked to a 3.8% increase in invalid voting. It is important to notice how in a linear model this effect also applies to negative variations, meaning that in years where protests fall in intensity, so do blank and spoiled ballots. This result shows that regardless of whether invalid voters actually show up at antigovernment protests, they are an integral part of a national context of political contention. This effect remains even after the interaction terms are introduced, except for the model that includes interactions with compulsory voting.

As for H2, the quantitative evidence from the statistical model contradicts the initial predictions, which saw labor mobilization as similar to antigovernment protests. The emergence of workers’ strikes seems to operate on invalid voting in a negative fashion, perhaps to signal that workers’ unions capable of mobilizing large numbers of protesters tend to be associated with the institutional politics of supporting large parties, not with protest voting. The Argentinian context is perhaps the most fitting for observing this kind of mechanism empirically in the positive, both after Peronism and since the 2001 crisis (De Riz 2008), while the end of workers’ strikes in early 1990s Bolivia also coincided with a fall in invalid voting. The two behaviors—antigovernment protest and workers’ strikes—therefore appear to be substitutes at the national level. The size of the effect was only slightly smaller than what was found for H1, just working in the opposite direction.

Concerning H3, a direct relationship between democratic deterioration and invalid voting is present with a large coefficient. In other words, the onset of more authoritarian politics prompts voters to protest by invalidating their preferences, while the number of valid votes for political parties increases in more democratic times. In model A, a one-point change in the democracy between two elections correlates on average with a 4.1% change in invalid voting of the same sign, all other factors being equal. This is an important result because it shows that democratic deterioration affects invalid voting directly, and independently from antigovernment rallies and workers’ strikes.

Conversely, the result for H4 clearly shows that at the aggregate level changes in political corruption do not have a direct impact on invalid voting in Latin America once other factors are held constant. Although we know from previous studies that corruption is undoubtedly related to protest voting at the individual level, after taking into account variations in democracy, protest participation, and workers’ strikes, the corruption variable is not statistically significant in any of the five regression models. One can still comment on the sign of the coefficient, which respects the general expectations in the hypothesis by presenting a positive relationship between corruption and invalid voting.

Finally, regarding the specifications that include interaction terms, model B shows that the corruption variable is still not statistically significant even after making it interact with changes in political protest, and does not mediate the relationship between protest and spoiled/blank ballots. Model C explains the largest amount of variation out of all models, with a coefficient of determination (R²) slightly above 30%. It also indicates that when public protests surge or wane, power alternation mediates their relationship with invalid voting. If the incumbent carries the election, the emergence of public protests at different intensities [1–4] corresponds to increases in invalid voting of 1.7–6.7%. When public protests wane under a lack of alternation in power, the negative relationship with invalid voting is similarly large. Consider instead that if the incumbent loses the election, protest surges are connected with negligible invalid voting increases of 0.1–0.4%.Footnote ⁶ The difference between the two scenarios is quite impressive and this key result indicates that Latin American citizens tend to participate in protests and recur to invalid voting when national politics see a lack of alternation.

Next, model D suggests a negative interaction between authoritarianization and public protests, even if the added term is not statistically significant. The sign indicates that the relationship between protest and invalid voting may be stronger when democratic conditions improve. Finally, model E contributes a surprising result, showing that compulsory voting leaves the relationship between changes in public protests and invalid voting unchanged. The coefficient for the adoption of compulsory voting is not statistically significant and constantly negative.

As for changes in the four institutional variables in the model, none of them yield any statistically significant results for invalid voting variations. Yet, it is possible to comment on the sign for changes in compulsory voting and the switch from concurrent elections to separate legislative and presidential elections. In both cases, the sign is negative, which confirms the insight from previous studies. Post-transition elections similarly have a slightly higher level of invalid voting, although the variation is not statistically significant. As for the economic variables, the change in the economic growth rate has a (surprisingly) positive sign and is significant across specifications. Still, previous studies showed that GDP growth in isolation is often ineffective in producing real improvements in Latin American lives (Janvry and Sadoulet Reference Janvry and Sadoulet2000). One can make a plausible argument that times of economic growth can increase the distance between politicians’ and citizens’ economic evaluations. In parallel, changes in national-level inflation yield no significant results in any of the models.