Contentious Electoral Discourse

Accusations of electoral fraud function as negative campaign statements for most citizens, but for a dedicated group who reward the candidate making them. This claim is counterintuitive for three reasons. First, democratization and electoral protest are not issues that candidates regularly use to gain public support, especially because democracy is not a good that candidates can provide once they gain office, but it is one that is provided to citizens by the system as a whole when credible elections get any candidate into office. In this way, democratization is not a good that can be provided once the conditions for a competitive, democratic election have been established; the good promised has been delivered already. In fact, democratization and the fight against electoral fraud are mainly expressive concerns for voters; they do not provide any direct material benefits to them (and material benefits can be easily provided by a willing authoritarian regime with sufficient funds). Second, and this is crucial, in a multiparty system weariness of electoral fraud should not be exclusive to one opposition group. Every party that suffers (or has suffered) from the disadvantages of being the victim of electoral fraud should be equally likely to denounce it, unless a party finds it instrumental to frame an electoral defeat as the result of a fraud, and other parties find that doing so is detrimental to their political strategy. Finally, not all citizens need to agree that they live under a democracy for it to squarely qualify as one. It is the rules of the game that define a democracy, and not what citizens think of them. Yet, the fact that some citizens may have a larger "democratic deficit" (^{Norris, 2011}) than others underscores the potential for political exploitation of this rift in public attitudes.

If an opposition party is able to successfully frame its defeat as the result of fraud, then it may be able to use the issue of the struggle against it as an issue to rally political support for coming elections. Collective action, especially for minority groups seeking social change, is fostered by the existence of common narratives that tie the efforts of all members together (^{McAdam et al., 1996}). The struggle against fraud can provide it, and it function as a flag to rally the support of its partisans. Expressive concerns like commitments to democracy (^{Riker and Ordeshook, 1968}) or the importance of the vote (^{Przeworski, 2010}) are part of the narrative repertoire available to candidates to attract the attention of voters (^{Vavreck, 2009}). Consequently, parties have incentives to link their brand to these issues if a relevant enough segment is interested in them. Expressive concerns become politically instrumental (^{Chong, 1996}) because parties get rewarded by attentive partisans for following up on their issues during a campaign. Once the association between the issue positions and the party identification is established, it is possible to say that the issue is "owned" by the party (^{Petrocik, 1996}) and that it has a strong incentive to keep it alive in its partisan's minds. The issue of democratization and electoral fraud does not need to be based on abstract notions of the need of a political reform, which have not been found to have an electoral impact in Mexico anyway, but it may be operated instead through the symbolic (and less information-demanding) mechanisms of an "easy issue" (^{Carmines and Stimson, 1980}) that can bring to mind the indignation produced by witnessing their candidate lose an election by means that, to their eyes, were illegal.

The Political Incentives For Contentious Electoral Discourse

The incentives for contentious political participation in a young democracy like Mexico are structured by its past as a competitive authoritarian regime, in which the electoral arena was heavily biased against challengers. ^{Levitsky and Way (2010, p. 5)} define a competitive authoritarian regime as one "in which formal democratic institutions exist and are widely viewed as the primary means of gaining power, but in which the incumbent's abuse of the State places them at significant advantage vis-à-vis their opponents". Elections happen, but the incumbent is able to maintain a strong foothold on the levers of government, by either flouting civil liberties, tampering with electoral procedures or giving itself significant material and communicational advantages. Levitsky and Way argue that power imbalances in the electoral arena push opposition strategies that increasingly seek to assure their survival through the use of protest, boycott or even violence, disregarding the maximization of their vote share. Yet some of these strategies are vote maximizing, if the conditions are right for a party to "own" them. This does not mean that electoral protest and accusations of fraud are the machinations of "anti-system" (^{Sartori, 1992} [1976]) or "semi-loyal" (^{Linz, 1978}) parties that aim to bring the regime down for political advantage. On the contrary, electoral protest and accusations of fraud are instantiations of a demand for more democracy. What they have in common with the former is their occurrence at the edges of democracy, where the institutional order is in a state of flux and it becomes relevant to elites and voters alike.

Electoral protest in competitive authoritarian regimes has, therefore, an ambiguous status. It can be interpreted as a call for mobilization in favor of democracy (^{Schedler, 2009}), as a failure of coordination among contending elites (^{Beaulieu, 2014}) or, given that opposition candidates who promote protests stand to gain access to power if they win (and their victory is recognized by the authoritarian incumbent), they can be interpreted as either the promotion of democratization or the promotion of a candidate (^{Schedler, 2014}).

To be sure, Mexico qualified as a competitive authoritarian regime until rather recent times. The Partido Revolucionario Institucional (Party of Institutional Revolution, pri) was formed in 1929 after the Mexican Revolution under the name of the Partido Nacional Revolucionario (National Revolutionary Party), a coalition of politicians that evolved to form the hegemonic party (^{Sartori, 1992} [1976]) that would rule the country for decades. It used its "hyperincumbency advantage" (^{Greene, 2007}), based on government patronage, spending, and selective use of violence and fraud, to remain in power. According to Levitsky and Way, Mexico became a full-fledged democracy after the 1996 electoral reform that severed the formal link between the executive and the IFE (Spanish acronym, Federal Electoral Institute). Further evidence of the democratic credentials of the regime came in 2000, when thePRIlost the office of the President to the Partido Acción Nacional (Party of National Action, PAN) for the first time in its history. This, however, does not mean that electoral contention would lose any of its power to mobilize voters.

On the contrary, it became part of the repertoire of contention (^{Tilly and Tarrow, 2007}) available to opposition parties, but used mostly by the Partido de la Revolución Democrática (Party of Democratic Revolution, PRD) (^{Eisenstadt and Poiré, 2006}). The use of electoral contention as a tool for mobilization depends, as does any other strategy, on the belief that it would prove to be successful. Such opportunity structure (^{Tarrow, 1998}) was established in the 1988 electoral scandal. By the time the 1988 election came around, Carlos Salinas de Gortari was chosen to be the next candidate of the PRI. Cuauhtémoc Cárdenas, a former governor, quit the party in protest and formed the Frente Democrático Nacional (National Democratic Front, FDN) together with smaller left-leaning parties and social organizations, and ran the first credible campaign against the PRI. This coalition was held together by the promise of patronage, Cárdenas' charisma, and his allegiance to the traditional tenets of the Mexican Revolution (^{Bruhn, 1997}). The 1988 election proved to be extremely controversial, no less because Cárdenas came close to winning. Allegations of rampant fraud were made, fostered by the deliberate crash of the early results system, taken offline by the government-appointed overseers of the election when results from urban precincts began to flow in favor of Cárdenas. The official tally gave Salinas 50 per cent of the votes, while Cárdenas got close to a third of them; many still consider the election to be the result of a large-scale fraud, however, it is worth mentioning that they resemble the results of pre-electoral surveys that were available to the public at the time (^{Domínguez and McCann, 1998}).

Manuel Clouthier, the candidate for the pan, was the first politician to cry foul. He later joined Cárdenas in massive protests against the alleged fraud, only to enter negotiations with Salinas when it became apparent that Cárdenas would be the only one to benefit (^{Bruhn 1997}). Moreover, his constituency seemed averse to protest, and the pan leadership believed that it would depress turnout and support (^{Loaeza, 1999, p. 368}). Regardless of whether panistas found accusations of fraud believable, or a reason to take to the streets, their leadership discounted this and therefore dismissed electoral protest as a tool for mobilization. Even if the pan had been victimized by fraud as much as the FDN, its leadership never used this situation to mobilize their bases, and consequently they never responded to it as a fraud. On the other hand, the magnitude of the scandal that resulted from his defeat gave Cárdenas and the FDN ample opportunity to use fraud as a rallying cry for his partisans to mobilize around, especially after the Frente broke up and the Party of Democratic Revolution (PRD) was created in early 1989 with some of its former members, a coalition that needed a new the "master frame" (^{McAdam et al., 1996}) to rally around.

Repertoires of contention are copied and adapted to new circumstances (^{Tilly and Tarrow, 2007}), in part due to their path dependency (^{Pierson, 2004}): the use of electoral contention as a strategy of mobilization features a positive feedback loop brought by the need of elites to find a master frame, i.e. the fight for democratization, and their partisans who want democratization to happen and their opposition candidate to win the election. Opposition parties in Mexico have used the call for democratization as a mechanism for mobilization, talking more about what will go wrong with the election before it happens, even as elections have become more credible and competitive (^{Vázquez del Mercado, 2013}). Some of the voters who support opposition parties will respond positively and support the candidate who makes the accusations (closing the feedback loop), while others will not. This interaction between party strategies and citizen demands is how issues evolve (^{Carmines and Stimson, 1989}), in particular the "regime issue" (^{McCann, 2012}) that the PRD has successfully exploited since its creation.

As an example of the adoption of accusations of fraud as contentious discursive strategies, when the 1994 election came around, the recently upgraded voters' registry was heavily criticized by the PRD for including duplicate and false names. The Salinas government had become so invested in legitimizing the elections that it organized a second debate, in addition to the first-ever televised one between the candidates, where the director of the Instituto Federal Electoral (National Registry of Voters, IFE) discussed the virtues of all the safeguards put in place with Samuel del Villar, an attorney and representative of the PRD (^{García and Figueiras, 2006, p. 118}). Six years later, the victory of Vicente Fox, from the pan, against the PRI in 2000 seemed to inoculate the political environment against contention, but the democratic honeymoon did not last long. As the 2006 election approached, the PRD complained that the pan and the PRI had maneuvered to reduce its representation in the governing council of the IFE. Perredistas were already primed to believe that another fraud was coming, and on election night the results were far too narrow for the election to be called. Felipe Calderón, of the pan, won by a margin of 0.58 per cent, and Andrés Manuel López Obrador (the PRD candidate) denounced that a fraud had been committed against him, demanding a full recount of the votes; the issue of whether Calderón had "stolen" the election proved very divisive. The stage was set for a full display of the power of accusations of fraud to mobilize perredistas around the flag waved by López Obrador, especially because, by then, it was the response that was expected of him (^{Bruhn, 2009}).

The Relationship Between Beliefs About Fraud and Protest Behavior

^{Norris et al. (2015)} have observed that the prevalence of contentious elections follows an inverted u-shaped curve when plotted against the level of democratization of the country. The peak occurs when competitive authoritarianism regimes celebrate elections under blatantly uneven circumstances, continually breaking the rules and giving opposition parties ample reasons to distrust electoral institutions and take to the streets when they lose. Mexican opposition parties suffered as much for decades, and even though Mexico no longer qualifies as an electoral authoritarian regime but as a full fledged democracy, Norris et al. show that its level of electoral contention remains high; its level of electoral contention approaches that of fellow young democracies like Kenya, the Ukraine, and the Philippines (p. 143).

Mistrust in electoral procedures is the attitudinal antecedent of contentious electoral behavior. Citizens of electoral authoritarian regimes and young democracies who think that elections were unfair, or even witnessed events of electoral malfeasance, are more prone to support protest activities or to participate in them. ^{Sedziaka and Rose (2015)} found as much in their research on the attitudes of Russian citizens after the 2011 State Duma elections, which were heavily criticized for their instances of manipulation. According to their survey results, respondents who believed that the election was manipulated were more likely to support electoral contentious behavior. ^{Norris (2014)}, using the results of the sixth wave of the World Values Survey, concluded that citizens around the world who feel that elections are manipulated feel much less confident about their electoral institutions, support their democracy less and participate in contentious behavior more often. Particularly, the perceptions of Mexican citizens about the integrity of their elections are more negative than the assessments made by experts (p. 107), indicating that there is a gap in public confidence in elections that is yet to be closed. Moreover, the relationship between attitudes about electoral integrity and their potential to trigger contentious behavior is structured by past electoral choices. Namely, losing elections makes citizens less trusting in democracy; this effect is even more intense for citizens who voted for parties that have been out of power for a longer time, or whose party of choice has never attained higher office (^{Anderson et al., 2005}). In particular, voting for a candidate who lost an election greatly increases a person's "protest potential," this is, their intentions to participate in contentious behavior like protest marches or attending public meetings organized to protest the government. This effect was larger for citizens in younger democracies (^{Anderson and Mendes, 2005}).

Furthermore, there are indications that concerns about electoral integrity (and their behavioral correlates) are not held equally among all citizens, even those who have lost. Schedler rightly postulates that "to the extent that citizens value democratic political goods and perceive the existing political regime to violate democratic precepts [...] manipulation makes democratic voters turn their backs on the government" (2009, p. 184). Yet, we need to better specify who these voters are and how elites are able to communicate with them. Especially because the evidence at hand paints a much less romantic picture of Mexican voters, who seem less concerned with democracy and much more with government performance. There has been no significant evidence (or rather, an abundance of negative results) in the Mexico Panel Studies from 2000 and 2006 (^{Magaloni and Poiré, 2004}; ^{Moreno, 2009}) of any link between partisan choice and considerations about whether political reform is necessary to guarentee a fair electoral competition. ^{Zechmeister (2008)} found that the relationship between citizen issue preferences and candidate positions related to political reform had no effect on electoral behavior. In short, the high-minded notion of political reform for a more equitable electoral competition failed to interest voters in a partisan manner, with the exception of voters who felt that the most important problem in the country before the 2000 election was political, who had a significantly higher probability to vote for the PRD (^{Magaloni and Poiré, 2004, p. 308}).

To summarize, believing that elections were stolen promotes contentious political behavior, and these beliefs are more prevalent, in general, for citizens who voted for those who lost an election. In particular for the Mexican case, they are more relevant for political behavior for perredistas because the PRD managed to "own" the issue of electoral fraud.

The Psychology of Accusations of Fraud and the Incentives to Make Them

A candidate who loses an election can choose to frame her defeat as the result of a fraud, an account that highlights some features of the election (e.g., the violation of regulations) at the expense of others (e.g., the candidate failed to mobilize enough support). By selecting which features of an event to emphasize and which to ignore, elites engage in framing effects that allow them, at least in principle, to influence the process of attitude formation in their relevant audiences. The prevalence of these effects in elite communication has led many students of public opinion to wonder about the ability of citizens to form correct and accurate opinions in the face of elite attempts to influence them through the information they choose to convey (^{Zaller, 1992}), yet framing effects depend both on the elite communication strategies (e.g., which features to highlight or which media or sources to use as channels of communication), and on the characteristics of the individuals who receive them (^{Druckman, 2001}). This has important implications for the study of the effects of accusations of fraud on the evaluation of the accuser, because her attempts to frame her defeat as the result of malfeasance by highlighting the failures of the electoral procedures that led to it will not be necessarily believed by everyone who hears them.

Frames operate by influencing which considerations come to mind at the moment of the formation of an attitude, and their "strength" depends on their availability in mind precisely when an opinion is formed, as well as on their applicability to the object being evaluated. Their availability depends on prior exposure to the frames, which become stronger with chronic exposure. And their applicability depends on the logical argument behind the frame, the credibility of the source, and other features (^{Chong and Druckman, 2007}). This means that elites who frame their messages in a certain way do not have complete control over which considerations effectively do come to mind and the effects they might have on attitude formation. In particular, their dependence on the context in which citizens have formed their predispositions and other features that determine the availability and applicability of frames determines their "strength", for instance, they can cause some frames to be "weak" and to have little or no discernible effect on attitudes. Additionally, if they tap into existing considerations that have a negative effect on attitudes, they can backfire and have an effect opposite to what was intended. As the effect of frames depends on context and predispositions, and these in turn vary across individuals, they will not affect attitudes in the same way for everyone; the same frame can have a strong effect for one person and leave another unaffected.

Elites influence attitude formation by using cognitive shortcuts to their advantage. The use of these cognitive shortcuts (^{Popkin, 1994}) amounts to a process of "satisfying rationality" (^{Simon, 1985}) that allows them to acquire and process small amounts of information and generate opinions that are "semi-automatic responses" (^{Lau and Redlawsk, 2006, p. 13}) rather than deep calculations of causes and consequences. In essence, this is one of the main tools for political communication, to "use rhetoric to trigger the psychological mechanisms that distort judgment" (^{Kuklinski and Quirk, 2000, p. 168}). But these triggers are not homogeneous among all citizens. Instead, they are mediated by political identity and social circumstance. That is what gives them their power to mobilize them selectively (see ^{Hillygus and Shields, 2009}). For instance, regardless of the current state of electoral institutions, elite complaints about electoral fraud can be used as a shortcut by a given group of voters (for whom the regime still seems quite prone to stealing elections) to determine the legitimacy of an election. Likewise, the same accusations may mean for another group of voters (whose candidate won the election) that the accuser is a sore loser who is complaining about the probity of the results because she has nothing else to say short of conceding defeat. Yet another group of voters (who did not turn out to vote because they are not usually interested in politics) may pay no heed to the accusations either because fraud is a thing to be expected or because it is not a thing that concerns them. Notice that these effects are exerted by the same stimulus. The difference in their reactions hinges on their predispositions; the most powerful and pervasive among them is party identification.

Party identification is a self-conception that, like religious identification, is strongly influenced by early socializations and comes with its own "doctrinal positions" advocated by other partisans (^{Green et al., 2002, p. 4)}. Early conceptions of voter identification (^{Campbell, et al. (1980} [1960]) shared the notion that it hardly changed at all, shifting slightly with changes in socioeconomic status but generally resisting political events, only coloring how citizens viewed them. Decades of research have shown a more complex picture, that insists on the solidity of partisan attachments but also on their interaction with events and information in the production of attitudes and choice (^{Zaller, 1992}). It influences the perception of political events and the evaluation of political actors, and even colors the evaluation of electoral institutions. For instance, ^{Green et al. (2002)} cite survey results about the events surrounding the disputed presidential election in 2000 in the United States which indicate that voters evaluated the legality of the procedures according to who they had voted for. A history of electoral abuse and repeated defeat would only worsen the distrust of losers. But their distrust is initially based on the ambiguity if the results and on the complexity of the procedures. Instead of discerning. Instead of discerning whether the elections were legal or the accusations have merit, citizens are better off falling back on their party identifications as heuristics to decide who won and how the decision was arrived at (^{Doherty and Wolak, 2012}).

Elites just cannot choose any issue of their liking and immediately link it to the identity of its partisans, especially if the relevant predispositions are not already in place. It is one thing for elites to mobilize support by appealing to issue positions, prejudices or other considerations, and another, much harder one, to mobilize them using an issue that is associated with the way individuals identify with a certain party, and what it means for them to identify with it. Does this mean that a party has to exist before the issue is linked with it? Not necessarily: for instance, if the issue of electoral fraud was experienced by the constituent organizations that would later become the PRD, then the issue of fraud would easily be transferred to perredista partisanship once the party was established. And it would remain in the minds of perredistas for as long as elites kept it salient in their minds.

It is relatively easy to mobilize support by appealing to issue positions, prejudices or other considerations. But it is much harder to do so using an issue that has already been associated to party identification. Parties want to create reputations to link themselves, in voter's minds, to policy or ideological positions. ^{Sniderman and Stiglitz (2012}) argue as much, supported by experimental results that show how partisans evaluate hypothetical candidates based not only on their partisanship, but also on the policies they advocate if they match the respondent's preference ordering. They show that respondents can provide the premium even without knowing the party of the candidate, as long as the ordering of the partisan options correspond to those of the voters. Respondents were able to infer their affiliation based on their policy proposals, and evaluate them accordingly by mapping their positions to what they know about how real-lIFE parties behaved, and gave the "premium" to their inferred co-partisans. This process provides an important methodological lesson. Respondents to an experiment who identify with a given party can locate a hypothetical candidate who gave a specific issue position and impute her a party membership, provided the party "owns" the issue. Then, they proceed to evaluate her. Such an experimental design avoids confounding evaluations caused by partisan identification of the candidate with whatever issues he stands for, while still allowing for the processes of evaluation related to respondents' partisan identification to take place. It will be used in the experiment presented below.

It is important to recall that the ability of parties to "own" an issue is related to the longevity of the party system. ^{Brader and Tucker (2009)} find that respondents to their survey experiments in Great Britain show a markedly higher tendency to increase the rating of policy proposals that are endorsed by the party they identify with, while this tendency is less evident for respondents from Hungary, and not observable at all in respondents from Poland. Hungary has had a more stable party system since the collapse of communism, and the system in Poland has been much less stable; the ability of the party brand to cue its identifiers about what policies to support suffers when it has had little time to develop and consolidate. These results attest to the necessity of party system stability for the consolidation of party identifications that not only allow voters to sort themselves, but also to allow them to use the party identity as a heuristic for attitude formation and choice. What about Mexico? The Mexican party system has been quite stable, the PRI leading the way with seven decades of existence and counting, while its leading contender, the pan, was created in 1939 but accrued enough support to, in latter decades, overtake the pri in presidential elections in 2000. The PRD was created in 1989, a two-decade existence that, according to the analysis of Hungarian results of Brader and Tucker, should be enough to create and sustain an identity. This does not mean that aggregate levels of party identification themselves have to be absolutely stable and unchanging. Green et al. make the case that party identification is stable as a social identity, but the party system in the United States is quite stable except for the major realignment that came as a consequence of the Civil Rights Movement in the 1960s. The vision put forth by ^{Fiorina (1981)}, according to which party identification works as a running tally of the current evaluations of the incumbent's performance, may be more appropriate for a regime where parties are younger and the rules of the game are still in flux, like the Mexican one. Although changing, it still retains its capability to anchor individuals' opinions and influence their behavior.¹

Expectations

It has been observed that citizens pay more attention to negative information (^{Soroka, 2014}), increasing the incentive of candidates to use it. However, there is a trade-off inherent in displaying a negative campaign. Candidates who embark on negative campaigns may be able to degrade the opinion of those they attack, but usually by paying a similar price themselves, at least in the opinions of the public at large, who may not find the negativity of the message acceptable (^{Fridkin and Kenny, 2004}, ^{Lau et al., 2007}). An accusation of fraud, especially one made during the actual campaign (either recalling past instances of fraud or denouncing current violations) constitutes a negative campaign because it is a denunciation of agents or institutions in charge of organizing the election; something inherently negative is being said about them. A first general expectation about the effect of accusations of fraud can be generated: an accusation of fraud will function as a negative campaign statement and will cause a drop in the evaluation of the accuser.

Frames that contain the accusations can vary in the features they contain and the effects they exert on the evaluation of the accuser. On the one hand, after decades under the "hyperincumbency advantage" of the PRI and some more reinforcement from recent scandals, it can hardly be expected for Mexicans to be indifferent to accusations of fraud. In general, attempts at electoral manipulation imply the failure to follow procedures to ensure legality and fairness of the election (^{Schedler, 2002}), which can affect the legitimacy of the result (^{Tyler, 2006}). Accusations of fraud have historically been framed emphasizing these failures, but they have become less frequent in past decades. For this reason, given the magnitude of the post-electoral scandal in 2006, attitudes related to it are expected to be more accessible when evaluating the accuser who reminds them of the scenario that brought it about (namely, a demand for a recount that follows a narrow election). An accusation can also have an effect due to its saliency in recent historical memory. A second expectation follows: an accusation based on the description of a procedural failure will have a smaller effect than an accusation that recalls a past scandal.

It is important to consider that, given that accusations will vary in their features, it will be necessary to control the grade of negativity of the message so that it can be perceived clearly. If accusations have an adverse effect on opinions about the accuser because they are negative, and not because of other features, then a third expectation is specified: a general measure of negativity will have a larger effect than either accusation. It is important to consider that the credibility of a frame depends in part on its context and, as a consequence, the credibility of an accusation of fraud depends on the electoral result. In particular, a narrow electoral margin can be interpreted as evidence of wrongdoing (^{Hartlyn and McCoy, 2006}). A fourth general expectation follows: electoral context, in the form of a narrow electoral margin, will give credibility to the accusation and reduce its negative effect.

The question remains about whether the effect of accusations of fraud is homogenous across the population. If it is, either because all citizens are indifferent or repulsed by them, then no candidate has an incentive to make them, other than promoting democratic development at the cost of "going negative". Conversely, if respondents from a particular group -who, in this case, identify with a particular party- respond to accusations of fraud as calls for mobilization, then they will respond to the accusations of fraud by improving their opinion of the accuser. A positive effect would say nothing about the candidates, or her partisans' commitment to the development of democracy. It would only say that accusations of fraud function as "wedge issues" (^{Hillygus and Shields 2009}), messages that appeal only to specific segments of an electorate and that can be used for the purpose of mobilizing them through mass communication. If indeed the PRD "owns" the issue of accusations of fraud and its partisans respond to it as a call for mobilization (but no other party does), then a fifth expectation can be formulated: respondents who identify with the PRD will reward the accuser, while respondents who identify with the other parties (or none at all) will reduce her rating.

Finally, to further probe the effects of recency of exposure to the frames, and therefore of their availability in mind, the effect of the age of respondents should be accounted for, and a sixth expectation can be formulated: "older respondents will recall the procedural failures more often and reduce the evaluation of the accuser who makes them". A complementary hypothesis is concerned with the capacity of political causes to mobilize young citizens who pursue their "self-realization" through participation in social movements (^{Rossi, 2009}) of having their candidate win the election and finally bring democracy about (after all, she only has to win). To the extent that personal expression and activism aimed at attaining democratization are preferred by younger citizens who prefer a candidate whose party has successfully positioned itself as a viable option for democratization, an alternative expectation can be specified: a candidate who makes an accusation of fraud will be evaluated more favorably by younger respondents who identify with her party.

Research Design

Citizens who voted for a candidate who lost an election have lower levels of trust in government (^{Anderson et al., 2005}). This behavior has the potential to induce a confounding effect between party identification and attitudes about electoral institutions associated with the response to accusations of fraud, such as trust that they will impede it. To minimize this, I propose to analyze accusations of fraud using a survey experiment in which respondents are read vignettes featuring a hypothetical election that varies in substantive and historically relevant ways, but in which the partisan identity of neither candidate is disclosed. This design has the drawback of minimizing the information available to respondents, but allows them to focus on the features of the accusation and the context surrounding it. Admittedly, this design affects the external validity of its results, especially because candidates are evaluated using other considerations (such as policy, performance or personality) with which the accusations would have to compete, but this is a small price to pay for the proof-of-principle that citizens respond to them in politically instrumental ways. The experiment discussed below was fielded in 2009, after a rather inconsequential midterm election, but following the very controversial 2006 presidential one, in which the debate centered on whether the pan (party of the president in office then) orchestrated a large scale fraud; this further minimizes the possibility that the results were driven by anti-PRI sentiment than by allegations of fraud.

Experimental conditions were designed to reflect historically occurring events. For instance, in the 1994 election, the opposition made continuous references to procedural failures like the existence of duplicated names in the register of voters. And in the 2006 post-electoral scandal, allegations were accompanied by a demand for a recount (made memorable by the chant of "vote by vote, station by station" or voto por voto, casilla por casilla). These two scenarios are featured in two different accusations: the first makes no reference to institutional failures but demands a recount, while the other does so explicitly; also, the former makes a reference to the most recent scandal, while the latter, to scandals more than a decade old. In order to maximize the credibility the accusations, and therefore construct validity of the experimental design, they were placed in time in relation the election. The "institutional" accusation is said to have happened before the election, and the demand for a recount is (necessarily) said to have happened after it. While this feature of the treatments follows the historical record as closely as possible, it has the effect of confounding the "institutional" nature of the accusation with its timing relative to the election, as well as its sequence in the historical record. It is quite possible that such difference affects responses in a way that the experimental design cannot control. A third, combined, treatment will be analyzed, which will be called "generalized negativity." It will be used to assess the effect of an accusation that obviates these distinctions, and will be composed as a variable that indicates whether the respondent heard either one of the accusations or heard both. The analysis of "generalized negativity" is intended to minimize this confounding effect by eliminating differences between accusations, at the same time that it will provide a summary measure of negativity that will allow the analysis to control for specific features of individual treatments.

The experiment had a 2x2x2x2 design, where respondents were read a randomly selected vignette that varied in the information it contained about the election² . It was introduced by asking the respondent to imagine a hypothetical election where two candidates competed:

Now, we are going to discuss the electoral process. Imagine that in an election for President of the Republic only two candidates competed: one from the party in government and another from the party in the opposition.

And then were read one, or a combination, of the following treatments:

In its absence, respondents were only read "IFE announced that the incumbent candidate had won the election" (absent any other treatment, this constitutes the vignette for the control group).

The dependent variable will be 100-point thermometer rating of the accuser (mean = 51.5, s.e. = 0.48). Further details about the dependent variable and the experiment are included in the appendix. Moderators will be tested by using sample splits. The results for the model, including only the basic treatments, and the full model (including "generalized negativity"), will be analyzed using OLS regression. The experiment was designed as follows: Prior to the actual experiment, the level of identification with the three major parties (PRI, PAN, and PRD) was measured. Each respondent was read only one randomly selected vignette and, immediately after, the thermometer scale ratings of the challenger were taken.

Given that the candidate is not identified by name or partisanship, respondents can "fill in the gaps" with whichever considerations they find applicable to her. For this reason, OLS regression will include a set of control variables related to attitudes that could affect the rating of the accuser, which could also be confused with the effect of the treatments. The first set of controls will measure levels of identification with the three main political parties (PAN, PRI, and PRD), as well as the opinion of their candidates in the presidential election of 2006. These controls will serve as indications of the affinity that respondents feel for the loser of an election, regardless of how she behaves after she is defeated. A measure of trust in the IFE will also be added to control for their legitimizing effect on the evaluation of the loser; to the extent that trust in the electoral arbiter leads to improved opinions of the competition and all candidates involved, higher levels of trust are expected to improve the rating of the loser. Finally, demographic variables like gender, age, and education level will be controlled for. Gender has been associated with political conservatism, in particular women have been observed to have a strong preference for the status quo and for incumbents (^{Moreno, 2003}), which leads to the expectation that they will also have a preference for winners and, conversely, be more critical of losers. For their part, due to their higher propensity to participate in non-institutional types of political activities (like signing petitions and joining protests) in an electoral context, younger citizens are expected to have a higher opinion of the loser (^{Norris, 2014}). Finally, education has long been observed to procure more critical attitudes towards electoral institutions and a higher demand for democracy (^{Norris, 2011}). Yet, it is not clear whether higher education could lead to improved attitudes about the loser of the election. The survey was fielded by the Unit of Applied and Opinion Studies, located in the IIJ-UNAM in September 2009.³ In total, 3 985 people were interviewed, selected from a nationally representative sample.

Results

Table 1 presents, in the left panel, the results of the OLS regression, including only the treatments and their interaction with "narrow electoral results." Other control variables are omitted from this table, but are available in the complete regression results in the appendix. The first striking fact to come out of these experiment is that, without the measurement of "generalized negativity" in the model, no treatment shows a significant effect. This is a result of the cross-effects induced by the inclusion of "narrow electoral results" in the model, which has two different results depending on what treatment it interacts with. These complex interactions are the result of the combined effects of the frame, its context, and the predispositions of the respondents, all of which can make framing effects difficult to predict (^{Chong and Druckman, 2007}). Once generalized negativity is included in the model, as well as its interaction with "narrow election," the results -shown in the right-hand panel- are much clearer. The effect of "generalized negativity" is significant and negative. By saying something negative about the election, the rating of the accuser drops by 6 points. This indicates that citizens do care about accusations of fraud, irrespective of their features, showing how they function as unwelcome negative campaign statements, and that citizens punish the candidate who makes them.⁴ Also of note is the interaction of "generalized negativity" with "narrow electoral results" has a positive effect. It is not statistically significant, which weakens its interpretation as the positive effect of a competitive electoral context on the credibility of the accusations, namely, to make it more credible and less prone to be interpreted as a negative campaign. Its lack of statistical significance indicates that the public does not believe an accusation because the context might support it, which has interesting implications for theories of contentious politics that seek to understand its effect on public opinion: the public in general does not appreciate contention much, even if it seems to be organized for a good reason; this also limits the ability of contentious groups to mobilize public support in their favor.

Table 1 Cost of negativity to the accuser. 

Source: Author's elaboration. Generalized negativity idicates the presence of either a pre-or a post-accusation or both. Entries are OLS coefficients, with robust standard errors in parentheses. Complete regression results in the appendix. ** -sig. at 0.01, * -sig. at 0.1.

This is particularly the case when observing the negative and significant effect of the combination of the demand for a recount with the narrow electoral margin, a scenario that closely resembles the 2006 scandal; it is brought out after removing the positive -but not significant- effect of its interaction with generalized negativity. Even three years after the scandal, Mexican citizens react strongly against it, showing how a frame can interact with its context because it is still available, and quite applicable, in the minds of citizens who evaluate the accuser demanding a recount. The Mexican electorate at large still remembered López Obrador and his protest. In general, it is possible to conclude that electoral contention of such magnitude is not easily forgotten by the public. Leaders of such movements who intend to take their causes to the public at large and expect their support should consider the negative long-term consequences on their reputation.

The effects of accusations of fraud are moderated by party identification. Table 2 contains the relevant coefficients for OLS regressions for above-median identifiers, for PRD, PRI, and PAN. Perredistas reward the accuser for making a credible accusation. They will significantly increase her rating if she makes either accusation of fraud -the effect of generalized negativity- when it is accompanied by a statement that the electoral results were narrow, by almost 31 points.⁵ The size of the effect is quite large, suggesting that not only do they mitigate the effect of negativity when hearing that the electoral result gives support to her accusation, but that they significantly reward her for making it. Furthermore, notice how it is the interaction of "generalized negativity" with "narrow electoral results" that has such a positive effect, and not the frame reminiscent of the 2006 scandal (which has no effect at all). This indicates that perredistas respond to negative talk about elections in general, when the electoral context lends it credibility, which hints at their continued demand for democratization. Likewise, the institutional accusation also causes perredistas to increase the rating of the challenger by 11 points (although this coefficient is significant at the .05 level for a one-tailed test). These results are a strong indication that the PRD "owns" the issue of democratization, to the extent that its partisans provide strong incentives to its candidates to make accusations of fraud. Democratization for perredistas is an "easy issue" that is linked more to their acquired distrust in elections and to their past struggle for democracy than to an actual assessment of the electoral results, then these results show that candidates from the PRD have easily activated this issue and reaped the benefits that come from mobilizing perredistas. At first, this would ring an alarm for those concerned about the stability of a democratic regime in which a loser can take her partisans to the streets to protest an electoral defeat. If the regime in question was under economic or social duress, protest would spill over onto other groups and cause damage to governability. This would be a scenario reminiscent to what ^{Sartori (1992)} describes as problematic in "plural polarized" regimes where anti-system parties are able to get close to bringing down the regime for political gain; only these partisans would not be rallying for less democracy but for more of it.

Table 2 Cost of negativity to the accuser moderated by party identification. 

Source: Author's elaboration. Generalized negativity indicates the presence of either a pre-or a post-accusation or both. Entries are OLS coefficients, with robust standard errors in parentheses. Complete regression results in the appendix. ** - sig. at .01, * - sig. at .1, + - sig. at .2.

However, perredistas participating in the experiment also show a negative reaction against accusations of fraud. In spite of their enthusiasm for a candidate who makes an accusation when the electoral context supports it, perredistas show a tendency to punish her when she "goes negative" on elections, and hearing that she lost in a narrow election causes them to punish her too. The main effect of generalized negativity indicates a 16-point drop in the rating of the accuser (though it is significant at the .1 level for a one-tailed test), just as the main effect of the narrow electoral margin causes a 12-point drop in her rating, perhaps as a reminder of the turbulent times.⁶ Together, these results run against the so-called culture of protest of perredistas, who seem quite ambivalent about accusations, the scandal, and the contentious events that accompany them (^{Estrada and Poiré, 2007}). In other words, the response (as of 2009) from perredistas to accusations of fraud comes from their response to the post-2006 fallout. These are clear limits to how much a candidate can do to mobilize her supporters, if they are interested in furthering democratization, before they start turning on her. And, in more general terms, social actors who are able to mobilize enough groups to protest can expect even their supporters to distance themselves from them. These results show that support for contentious discourse and politics has a self-limiting nature.

To complete the analysis of the effect of partisan identification on the reception of the accusations of fraud, an analysis of respondents who identify with the PRI, the pan, and with none of the parties was carried out. Supporting the claim that the PRD "owns" the issue of electoral fraud, neither priístas nor panistas responded in significant ways to the frames.⁷ Given that the level of party identification was measured independently for each party on a 5-point scale, independents were defined as those respondents who answered that they did not identify with any one of the parties listed (the three large ones, as well as four smaller ones that were omitted from the present analysis due to their small number of identifiers). Results, while significant at the .1 level for a one-tailed test, suggest that citizens who strongly reject identification with any party respond to accusations in a way that is similar to perredistas, who punish the loser for making an accusation, yet reward her for making a credible one. These similarities are important because they shed light on how the attitudes of citizens who identify with no party at all are similar, based on attitudes of distrust of parties and elections that make their response to accusations of fraud quite similar to citizens who have had experiences of electoral defeat. This gives accusations of fraud the potential to mobilize citizens beyond a party's habitual audience, perhaps deepening the incentives to make them.⁸ They also shed light on the incentives faced by independent candidates who may be able to mobilize voters by making negative-if credible-statements about elections.

Finally, to explore the possibility that frame strength depends on previous exposure to it, in particular the accusation related to repeated names in the voter's registry, the sample was split between those who would have been of voting age or older in 1994 and younger ones. Respondents who were 18 in 1994 would be 33 in 2009; as age was coded in 5-year groups, younger voters were identified as those who belonged to the categories of 30-34 years old or younger, while older ones belonged to the categories 35-39 years old or older. Table 3 contains these results. Against the expectation that the institutional accusation would be more credible to older respondents, younger ones proved to be much more sensitive to all of them, while older ones were not responsive to any; it is not possible to assert whether this is because youth is more easily impressed by electoral protest or because older voters have become insensitive to them, but perhaps it's a combination of both. Like perredistas, who show a barely significant tendency towards punishing the accuser, younger respondents are not tolerant towards negativity, dropping her rating by 16 points, but their reward for her going credibly negative has a smaller magnitude, increasing her rating by 20 points. Unlike perredistas, they do reward the accuser for demanding a recount by increasing her rating by 9 points, but punish her for demanding it after a narrow election by dropping her rating by 16 points. These apparently contradictory results are explained by the fact that the response of the young is also related to their party identification. Youth overall seems to resent the post-electoral scandal in 2006, punishing the accuser for demanding a recount after a competed election, while youth who also identify with the PRD do not. Notice, in the results in Table 3, that the reward that perredista youth give to the accuser who goes credibly negative is a 60-point increase in her rating, but they also take away 38 points when she goes negative. This is further evidence that perredistas, especially young ones who have more extreme reactions, have little patience for baseless accusations. Allegations of fraud as tools for mobilization are double-edged swords.

Table 3 Cost of negativity to the accuser moderated by age and identification with PRD. 

Source: Author's elaboration. Generalized negativity indicates the presence of either a pre-or a post-accusation or both. Entries are OLS coefficients, with robust standard errors in parentheses. Complete regression results in the appendix. ** - sig. at .01, * - sig. at .1, + - sig. at .2.

The coefficients of control variables (shown in the appendix) behave in interesting ways. Given that the effects of accusations are already controlled for, the effect of the remaining variables may be interpreted not as being exerted on the accuser, but merely on the loser of an election. They do not vary much across model specifications, and show how respondents who identify with both parties that lost in the 2006 election have a more positive opinion of the loser of the election: the coefficients are small and barely significant for príistas, but are much larger and significant for perredistas, which is not surprising given the high profile their candidate gained after the election. This result is supported by the effect of the evaluation of the candidates. Having a positive opinion of López Obrador has a statistically significant effect on the evaluation of the loser, while having a positive opinion of Roberto Madrazo (the PRI candidate who can against López Obrador) has an effect half of the size. Neither identification with the pan or opinion on Felipe Calderón have significant effects. On the other hand, the IFE has a legitimizing effect on the loser: increasing levels of trust in this institution are associated with higher ratings for her. This is true for all respondents but for perredistas, whose attitudes about the loser of the election worsen with increasing trust in the IFE. This last result shows the ambivalence that perredistas felt after the 2006 scandal, their attachment to electoral institutions being contravened by their candidate's attack on them. Regarding the demographic variables, gender showed a significant effect: women like a loser less than men. This may be related to the long-observed preference that women showed for the PRI. Also, older respondents have a slight tendency to dislike the loser. In view of the much stronger effect that accusations have on younger respondents, their higher evaluation of the loser is an indication that they may be paying more attention to what happens during elections and may be more amenable to political change. Finally, against expectations that education would lead to improved attitudes about the loser, no significant effect was observed across model specifications.

Discussion

These results should give pause to challengers interested in exploiting the potential of accusations of fraud to delegitimize winners: in general, voters do not take such accusations lightly and would rather believe their institutions work; they do not appreciate hearing about fraud. This result is quite probably related to the current state of Mexican democracy: its transition is finished and demand for democratization is now residual to some pockets of the population. This is important for the structure of costs and benefits experienced by the accuser, whose choice is either to make an accusation of fraud or accept the results. To the extent that considerations about electoral legitimacy dwell on the "regime dimension" of public opinion in a consolidating democracy, the effects of accusations on electoral choice cannot be neglected. A case in point is the so-called "culture of protest" of perredistas, whose perennial status as outsiders to the office of the president (and a leadership chronically eager to make accusations of fraud) have conditioned them to respond to negative talk about institutions as a "wedge issue." This strategy has proven to be very polarizing, and aggravating even for perredistas, who are seemingly becoming more skeptical about their candidates making wanton accusations of fraud.

By the 2012 campaign, the PRD seemed to have learned as much. Before the eruption of the #yosoy132 movement, López Obrador had been running a rather quiet operation, even in his first TV spots he "held out a hand, frankly, to whom I may have affected in my determination to fight for democracy and peace"⁹ and talking about the future he wanted for his infant son, who sat playing next to him and received a kiss on the forehead at the end. His tone shifted after the election, calling the process "unequal" (^{Nieto and León, 2012}) but demanding that no violence or large-scale protests should take place; then calling the elections a "national disgrace" because the IFE was not doing a recount (^{Nieto, 2012}). The process ended with riots outside of the Cámara de Diputados (House of Representatives) on December 1st, 2012, while the priísta Enrique Peña Nieto was being sworn into office. Both ^{Beaulieu (2014)} and ^{Norris (2014)} mention them as recent instances of conflict under admittedly ambiguous circumstances. There were indications that the rioters, who did not identify themselves with any party, were in fact associated with the PRD. By the end of December the perredista majority in the Asamblea Legislativa de la Ciudad de México (Legislative Assembly of Mexico City) amended a law that allowed people found guilty of disturbing the social peace (a remnant of an authoritarian past) to be set free on bail, and perredista members of the federal lower House of Congress paid to set the remaining 14 people free (^{Milenio Diario Online, 2012}). It seemed that the PRD had kept at its strategy of mobilizing voters by appealing to their considerations about the regime. Only this time, they had learned that "going negative" has a cost and tried to avoid it by playing a safe campaign, tapping into social movements already blossoming or fostering protest activities without openly being associated with them. The attitudes of perredistas observed here are resonant with this dual strategy. This behavior might be a sign of a more sophisticated strategy, one that adapts to changing times and public preferences. As democratic institutions gain credibility in the larger electorate, accusations of fraud and protest diminish their capacity of mobilization. The costs of "going negative" in elections increase with time, up to the point that only a few voters remain attentive to calls of protest. Their potential for mobilization is still there: right before the election, perredistas were more prone (52.4%) than panistas (33%) or priístas (29.1%) to believe that the elections would not be clean, and they still largely considered that the return of the PRI to the presidency was a problem for Mexican democracy (78.6%), more so than panistas (54.7%).¹⁰