Functional democracies require not only a fair electoral process, but also voters' belief that such is the case. The way in which citizens perceive the integrity of the election is a relevant source of legitimacy and political participation (Birch Reference Birch2010; Norris Reference Norris2014). Scholars have therefore explored the determinants of election trust, demonstrating the importance of institutional factors such as the impartiality of the electoral authorities, the efficiency of the voting process, and the power of voters' idiosyncratic characteristics. An important yet disturbing insight in this literature is that voters' confidence depends on subjective factors as much as on the objective conditions of the election.
The best illustration of how subjective factors shape electoral trust is the importance of the “winner/loser” status. The distrust among those who voted for a losing candidate and its contrasting effect among those who voted for the winner have been extensively documented across multiple elections (Alvarez, Hall, and Llewellyn Reference Alvarez, Hall and Llewellyn2008; Anderson et al. Reference Anderson2005; Maldonado and Seligson Reference Maldonado, Seligson, Norris, Frank and Martínez i Coma2014). However, it remains unclear whether such reactions reflect only the election result or also depend on the candidates' assessment of the process. Distinguishing both components allows us to evaluate the extent to which candidates' refusal to accept the electoral outcome can affect voters' ultimate confidence in the process.
To narrow this gap, this letter focuses on the relationship between election trust and the reaction to an election result by losing candidates. We expand on the recent works explaining the incentives for candidates and parties to challenge the election outcome (Beaulieu Reference Beaulieu2014a; Chernykh Reference Chernykh2014; Hartlyn, McCoy, and Mustillo Reference Hartlyn, McCoy and Mustillo2008; Hernández-Huerta Reference Hernández-Huerta2020; Lago and Martinez i Coma Reference Lago and Martinez i Coma2017), and explore the consequences of such challenges on voters' election trust. We argue that candidates disputing the electoral outcome widens the so-called ‘winner–loser gap’ in election trust. In particular, the effect of a candidate challenging an election outcome is more prevalent among voters on the losing side, for they are more likely to seek out and believe negative assessments of the election itself.
We test our prediction by combining the extensive Latin American Public Opinion Project (LAPOP) survey data with contextual information for almost 50 presidential elections in Latin America. Our main finding is that the negative effect of supporting a losing candidate on electoral trust is about twice as large when a losing candidate challenges the outcome. This relationship holds across different robustness checks and specifications to account for the integrity of the election and candidates' reactions to the electoral outcome.
The Catalyst for Election Distrust
The perceptions of electoral integrity are shaped by several objective and subjective factors. Most of the objective predictors focus on the institutional conditions that promote a “level playing field” (Birch Reference Birch2008). These conditions include, for example, the autonomy and performance of the electoral management bodies (EMBs) (Garnett Reference Garnett2019; Rosas Reference Rosas2010), along with the fairness of election laws (Frank and Martínez i Coma Reference Frank and Martínez i Coma2017), public funding for political parties (Birch Reference Birch2008), or the quality and quantity of information that voters receive (Kerr and Lührmann Reference Kerr and Lührmann2017). The common assumption of all these works is that voters are able to perceive the factual conditions of electoral integrity.
By contrast, the subjective predictors include those factors that affect voters' election trust even when they are not necessarily related to the integrity of the process (Daxecker, Di Salvatore, and Ruggeri Reference Daxecker, Di Salvatore and Ruggeri2019). For example, several studies have shown how partisan attachments color individuals' perceptions of vote fraud (Alvarez, Hall, and Llewellyn Reference Alvarez, Hall and Llewellyn2008; Ansolabehere and Persily Reference Ansolabehere and Persily2008; Beaulieu Reference Beaulieu2014b). Other factors include voters' political sophistication (Karp, Nai, and Norris Reference Karp, Nai and Norris2018), predispositions to believe conspiracy theories (Norris, Garnett, and Grömping Reference Norris, Garnett and Grömping2020), and preconceptions about news sources (Bush and Prather Reference Bush and Prather2017).
Perhaps the most-studied subjective predictor of election trust, and other indicators of trust in government, is the voter's “winner/loser” status (Anderson et al. Reference Anderson2005), that is, those who voted for a losing candidate tend to show lower levels of election trust than those who voted for the winning candidate. This response is rooted in individuals' emotional reactions to winning and losing (Brown and Dutton Reference Brown and Dutton1995). Additionally, voters for a losing candidate build a negative assessment of the integrity of the process to relieve the cognitive dissonance associated with facing an electoral defeat despite supporting the “best” candidate (Daniller and Mutz Reference Daniller and Mutz2019). A growing body of literature has also offered rich nuances about the effect of the “winner/loser” status on trust in elections, for instance, the gap in electoral trust between winners and losers is moderated when voters perceive higher levels of electoral integrity (Maldonado and Seligson Reference Maldonado, Seligson, Norris, Frank and Martínez i Coma2014; Mauk Reference Mauk2020) or when a non-partisan entity certifies that elections were well conducted (Kernell and Mullinix Reference Kernell and Mullinix2019). Furthermore, there is evidence of heterogeneous effects of partisanship among losers (Cantú and García-Ponce Reference Cantú and García-Ponce2015).
We argue that distrust among election losers intensifies when a candidate challenges the election result. This is the result of voters' limited capacity to monitor the integrity of the entire electoral process themselves. As a result, when assessing the integrity of the election, citizens often rely on personal experiences (Kerr Reference Kerr2018) or media coverage (Norris Reference Norris2014).
In particular, we expect that candidates' negative claims about the election will resonate more with voters on the losing side. Our expectation is based, first, on the fact that voters are more likely to seek out information from candidates and parties they support (Goren, Federico, and Kittilson Reference Goren, Federico and Kittilson2009; Robertson Reference Robertson2017). Additionally, theories of motivated reasoning suggest that voters will attribute different importance to the available information on the integrity of the election, depending on whether they are on the winning or losing side. In this case, motivated reasoners are more likely to search for evidence consistent with confirming information, regardless of the accuracy of the source (Kunda Reference Kunda1990; Little Reference Little2019). As a result, supporters of losing candidates are more attentive to their allegations of fraud and view the negative rhetoric about the integrity of the election as an acceptable explanation of the election defeat.
In sum, voters on the losing side are less trusting of the integrity of the election. The reason for that distrust has to do with the emotional and cognitive processing of losing. We expect this reaction to be inflated when a candidate rejects the outcome of an election. Voters on the losing side are motivated to consume and believe this type of information, which is used to construct a negative assessment of the election's integrity. Supporters of the winning candidate, on the other hand, are hardly affected by the rejection of election results or by fraud allegations, as they did not receive the directional signal from their preferred candidate. Therefore, our working hypothesis is as follows:
Hypothesis 1: The decline of election confidence among respondents on the losing side is magnified when a candidate challenges election results.
Empirical Strategy and Data
To study the relationship between voters' election trust and candidates' refusal to accept an outcome, we use data carried out by the LAPOP in 18 Latin American countries. These surveys include the answers of almost 100,000 individuals from 2004 to 2018. Our dependent variable, Election trust, uses answers on a scale of 1 (not at all) to 7 (a lot) to the question: “To what extent do you trust elections in this country?” The overall mean of this variable in our database is 4.17, and it ranges at the country-wave level from 2.46 in Paraguay in 2008 to 5.98 in Uruguay in 2010.
We explore election trust in relation to two, key independent variables. First, Vote loser identifies those respondents who voted for a losing candidate, using the answers provided in each survey to the question: “For whom did you vote for president [in the last presidential election of the country]?” Second, Result rejection is a dummy variable identifying a presidential election after which a runner-up candidate made a public statement rejecting the validity of the results. This variable covers all Latin American presidential elections from 2001 to 2018 and follows the same coding rules as Hernández-Huerta (Reference Hernández-Huerta2020). Our coding found instances of disputed outcomes in 24 per cent of the elections. The Online Appendix provides a detailed explanation for each of the elections considered in the database.
Our theoretical expectation is that among those who voted for a losing candidate, the value of Election trust should be lower when a losing candidate questions the integrity of the process. To test for this expectation, we model a cross-level interaction between Vote loser and Result rejection. The effect of this interaction is interpreted as the additional change in election trust among voters on the losing side of a disputed election.
We also include a battery of control variables that the literature associates with electoral trust and that are available in all our survey waves. At the individual level, we consider covariates such as the respondent's Age, Gender, and years of Education, as well as their self-reported Interpersonal trust. At the election level, our most important control addresses the possibility that distrust and candidates' reactions to the process indeed reflect an unfair election. Using data from the Varieties of Democracy project (Coppedge et al. Reference Coppedge2020), Free and fair (V-Dem) summarizes election experts' assessments of registration fraud, systematic irregularities, intimidation of the opposition, vote buying, and election violence. Other election-level variables include the Margin of victory and how long the country has been democratic by the time of the election (Years of democracy). Finally, we include a factor variable classifying the electoral rules for the presidential election as Plurality (our baseline category), Runoff, or Runoff held by Congress. The summary statistics and the coding scheme of all these variables are available in the Online Appendix.
Additional tests include a battery of control variables available only in a subset of surveys. Party identification follows Birch's (Reference Birch2008) approach and captures whether the respondent identifies with the party that won the election (reference group), with any of the losing parties, or with no party. Days after the election is the log number of days between the election day and the interview date. This variable addresses the possibility that respondents' ability to recall their vote and assessment of the election decays over long periods of time. We also include two important variables related to levels of electoral trust: Political interest, Ideology, and a quadratic transformation of the latter to model the effect of ideological extremism.
We specify a linear multilevel model with voters nested within elections and elections nested within countries.Footnote 1 This approach allows us to model different random intercepts for each election and country, accounting for the omitted covariates at both levels that may affect the levels of election trust across respondents.
Results
Table 1 summarizes the main results and some of our robustness checks.Footnote 2 The findings for our benchmark estimation are in column 1, which includes the interaction of Vote loser and Result rejection plus the main control variables at the individual and election levels. The coefficient for Vote loser confirms the lower levels of election trust among those voters on the losing side. In this case, the average difference in Electoral trust between those who voted for the winning candidate and those who supported someone else is about 0.4 points. Such distrust among those who supported a losing candidate is amplified under a disputed election. In particular, as the interaction between Vote loser and Result rejection shows, the election trust among those who supported a losing candidate is about 0.6 points lower when the election outcome is disputed than when it is not.Footnote 3
Determinants of election trust
Notes: .All estimates are based on ordinary least squares regressions. Standard errors are shown in parentheses. Main control variables include Female, Age, Education, Interpersonal trust, Years of democracy, Margin of victory for the election, and the Electoral rule for choosing the president. Additional control variables include Political interest, Political ideology, Party identification, and Days after the election. V-Dem refers to the Varieties of Democracy project, NELDA refers to the National Elections Across Democracy and Autocracy data, and PEI refers to the Perceptions of Electoral Integrity database. *** is significant at 0.1 per cent; ** p is significant at 1 per cent; and ** is significant at 5 per cent.
To interpret the results more substantively, Figure 1 presents the average predicted values of our dependent variable across supporters of winning and losing candidates in challenged and non-challenged elections, leaving all other variables at their median levels. As Figure 1 shows, supporters of the winning candidate display similar levels of trust in elections, irrespective of whether the election outcome was challenged or not. By contrast, the winner–loser gap when the runner-up candidate rejects the outcome is at least twice as large as what is observed when candidates conceded defeat.
Estimated effects.
Notes: The figure shows the estimated effects on Electoral trust by: (1) whether the respondent voted for a winning or losing presidential candidate in the last election; and (2) whether the electoral result was challenged by one of the candidates. Each bubble shows the average value for Electoral trust by election and voter type. Triangles are the median estimated effect and the vertical lines denote the 95 per cent confidence interval for each voter and election type.
A suggestive way to confirm this relationship is to compare the values for our dependent variable in Mexico's presidential elections of 2012 and 2018, where, despite the consistency of the institutional and logistic conditions for the elections, a candidate who ran in both elections had a different assessment of each event. In 2012, the election was plagued with informal and legal challenges to the election result by the runner-up candidate, Andrés Manuel López Obrador. For this election, the average values of Election trust for voters on the winning and losing side are 3.8 and 2.9, respectively—a difference of 0.9 on our seven-point scale. In 2018, López Obrador won the election and all the losing candidates conceded defeat. In this case, the average values of the dependent variable are 4.2 and 3.8 for election winners and losers—a difference of only 0.4 points. We corroborate this effect using panel data from 2012 and 2018 and present the results in Section C in the Online Appendix.
Columns 2–6 in Table 1 summarize the results of some of our robustness checks included in the Online Appendix. The goal of these models is to verify whether our results hold after considering additional controls and alternative codings of the dependent and main independent variables. Column 2 replicates the analysis, including Party identification, Ideology, Political interest, and Days after the election. While these variables are available only for a subset of surveys, the results follow what is expected by the literature (see Model 2 in Table A2 in the Online Appendix). Partisan losers and those with no partisan identification display lower levels of trust than partisan winners. Similarly, respondents report lower levels of Election trust when they have less interest in politics and are interviewed more days after the election. In any case, the magnitude and statistical significance of our variables of interest show similar results to those of our main specification.
Columns 3 and 4 test for the robustness of the results by using alternative measurements for an election challenged by a candidate. In particular, we consider whether they proceeded legally in court against the integrity of the election (Legal challenge) or led a post-electoral protest against the very integrity of the process (Post-electoral protest). Both models show that the main findings remain unchanged under these specifications.
Table A3 in the Online Appendix also verifies the robustness of the findings with two alternative measurements of Result rejection. The first one is an additive index for whether a candidate makes a negative claim, proceeded legally in court, and led a post-electoral protest (see Model 3, Table A3 in the Online Appendix). The second one uses data from the Perceptions of Electoral Integrity (PEI) survey (Norris and Grömping Reference Norris and Grömping2019) to build Result rejection (PEI), an election-mean score of experts' agreement, on a scale of 1 (“strongly disagree”) to 5 (“strongly agree”), to the statement: “Parties’ candidates challenged the result.” Both robustness checks show very similar results to our main model. Also, we included year fixed effects (see Table A7 in the Online Appendix) to control for factors changing yearly and constantly across countries, such as possible trends in levels of election trust. Our results remain unchanged under any of these specifications.
We also acknowledge that our coding for Result rejection captures only the behavior of the runner-up candidate. We expect that adding the reaction of more candidates will increase voters' distrust in the election. Therefore, if anything, our variable is a noisy measurement of the heuristics that voters receive about the integrity of the election.Footnote 4 To explore the specific effect of the challenging candidates upon their supporters, Model 3 in Table A2 in the Online Appendix replicates the main analysis, substituting Vote loser for Vote first loser and Vote other loser, which identifies those voters supporting the runner-up in the election and other losing candidates, respectively. As our results show, both sets of supporters of losing candidates display lower levels of trust in elections, and this effect is magnified for both groups when interacted with Result rejection. Nonetheless, the magnifying effect is stronger among the supporters of the runner-up candidate than among the supporters of other losing candidates.
Finally, Columns 5 and 6 in Table 1 verify whether our main findings hold under alternative ways of measuring the integrity of the election. The first one uses data from the National Elections Across Democracy and Autocracy (NELDA) dataset (Hyde and Marinov Reference Hyde and Marinov2012) regarding whether there is “evidence of domestic or international concerns that the election process was not going to be free or fair.” The second one uses data from the PEI database to summarize experts' answers to 49 substantive variables regarding the compliance of the election with international standards. An additional test in Table A4 in the Online Appendix uses data from the V-Dem project, and instead of directly controlling for electoral integrity, controls for the autonomy of the EMBs from the government. Our results are robust under any of these specifications. An additional test splits our sample into elections with high and low levels of electoral integrity (see Table A5 in the Online Appendix). While the results remain significant for both subsamples, the magnitudes of the coefficients for Vote loser and Result rejection are larger among elections with low levels of electoral integrity, confirming the moderating effects of electoral integrity on electoral trust (Maldonado and Seligson Reference Maldonado, Seligson, Norris, Frank and Martínez i Coma2014; Mauk Reference Mauk2020). In other words, voters assign more credibility to candidates' rejection of election results in contexts where violations of electoral integrity are feasible.
Discussion
The idea that supporters of a losing candidate have lower levels of trust in elections is a well-established finding in the comparative politics literature. However, as Anderson et al. (Reference Anderson2005, 142) note, there are “standing differences across different kinds of losers across different countries.” In this letter, we differentiate between elections in which losers conceded defeat and those in which they challenged the election. We argue that when a losing candidate refuses to accept an election outcome, distrust in elections is magnified among the loser's supporters.
One way to interpret our finding is that although the negative effect of supporting the losing candidate has often been explained as an affective response of losers' supporters, part of this effect is also driven by the actions of the losing candidates themselves. In particular, when the supporter of a losing candidate receives information that their preferred candidate refuses to concede the election, that individual is directed in motivation by this action and uses the information received to construct a negative image of the election, leading them to distrust it. This result is robust to other controls, including the respondent's partisan identification and indicators of the overall quality of the election.
This research has implications for scholars and practitioners of elections alike. It suggests that objective variables, such as the degree to which an election is free and fair, are not necessarily the main determinant explaining trust in elections, especially when the losing candidate refuses to concede. The information conveyed by the behavior of losing candidates is a strong cue assimilated by their supporters, which, in turn, could magnify possible irregularities that might have been present during the election. Therefore, the efforts of electoral authorities to improve trust in elections should not be circumscribed to improving electoral integrity itself, but rather be accompanied by campaigns which emphasize that in democracies, it is expected that losers accept the election results.
Supplementary Material
Online appendices are available at: https://doi.org/10.1017/S0007123421000399
Data Availability Statement
Data replication sets are available at bit.ly/PublicDistrust and in the Harvard Dataverse at: https://doi.org/10.7910/DVN/CAFHEB
Acknowledgments
We thank two anonymous referees, Abraham Aldama, Rodrigo Castro-Cornejo, Don Lee, Sandra Ley, Arturo Maldonado, Miguel Sánchez and David Sulmont for their useful comments. This research also benefited from feedback at the CIDE's Política y Gobierno Seminar, the University of Houston's Political Science Brown Bag Seminar, and LAPOP's Public and Democracy in the Americas conference.
Financial Support
None.
Competing Interests
None.
