Introduction

As the negative tone of elections increases (^{Geer, 2012}), it could be assumed that this trait is also growing in the advertising produced by the campaigns themselves; after all, if the discourse of the candidates revolve around attack, aggression and insult, political advertising might follow suit. The increase of negativity in the American campaign of 2016 and particularly the political advertising of 2012 of said country (^{Fowler & Ridout, 2012}; ^{Hill et al., 2015}; ^{Marland, 2015}) could signal the arrival of an era centered on this tone of the messages. Scholar’s concern about them are the negative effects that have been found in some circumstances, such as the decrease of voter turnout or political cynicism (^{vanHeerde-Hudson, 2011}; ^{Walter, 2014b}) and its relation to other issues, like polarization, disaffection, or public opinion silos in social media (^{Dahlgren, 2015}).

Nonetheless, these findings do not consider that any message is linked to the political, material, and regulatory conditions where they are produced and that shape their treats (^{Murdock & Golding, 2000}). This means that the particular regulation of political advertising and the structural conditions of the political systems have consequences on its frequency and the tactics deployed by them. Empirical research demonstrates these effects: the quantity and traits of negative advertising are explained by the fact that the attacker is either a challenger or an incumbent (^{Salmond, 2011}; ^{Walter, 2013}, ^2014b), the level of the office in dispute -executive or legislative, federal or state-, the party profile -emergent or traditional-, the timing of the elections (^{Sullivan & Sapir, 2012}) and the degree of polarization and ideological proximity of the rivals (^{Geer, 2012}). In light of electoral governance, this is an important issue: if negative campaigning is not a homogenous and inevitable trend, but is subject to external political conditions, then there is room for policies to correct them (Walter, 2011).

As a case study, Mexico has many interesting features. Is a nation recently transitioned to democracy, in 2000, though the attention to negative political advertising emerged in the 2006 elections, with its unprecedented strong tone, high frequency, and the tightness of the final tally. Such circumstances gave way to a regulation that sought to eschew it from the campaigns by defining it more precisely and raising the sanction in case of wrongdoings (^{Valdés, 2015}). Empirical research confirmed that such advertising was indeed prone to slander (^{Gutiérrez, 2007}), though it did not support the fact that was as pervasive as the media and politicians contended (^{Freidenberg & González Tule, 2009}; ^{Juárez, 2007}). On the other hand, their consequences are contrasting, rather negative in the institutional dimension and ambiguous in the electoral one, in terms of their effects, sometimes positive, others negative and others innocuous (^{Guerrero & Arellano, 2012}; ^{Sánchez & Aceves, 2008}). Nonetheless, from that election on, a few studies have explored such issues (see ^{Díaz & Alva, 2016}, for an exception on the matter).

This article aims to explain that the frequency of attacks in political advertising, as well as their tactics (how they do it), are a function of political factors and campaign situations. Our general hypothesis is that the characteristics of the content are shaped by factors outside of the will of the candidates. The cases analyzed, negative ads, come from the latest four Mexican presidential elections (2000, 2006, 2012 y 2018). Because of their longitudinal quality and breadth of cases, we can statistically test the previous assumptions. The theoretical frame explains the political and regulatory factors that shape political campaigns, as well as the main hypotheses and variables pertaining to negative political advertising. Findings support or refute those hypotheses.

Negative advertising. Characteristics and explanatory factors

Attacks between opponents is not a novel campaign practice. Throughout history, in ancient democracies such as the Athenian, and modern ones such as the American, negative slant, verbal aggression and incivility in campaigns have been amply recorded. Sometimes they tend to increase, raising worries among critics as a sign of decadence, though most of the time they are seen as a natural component of the competition for power (^{Richardson, 2001}). What was innovative and sometimes worrisome from the second half of the twentieth century and on is the possibility to advertise political messages on television, since the wider audiences at its disposal and emotional appeals that are capable of, could have a greater impact than ever before.

Political advertising is defined as “any message controlled through any channel, designed to promote the political interests of individuals, parties, governments and other organizations” (^{Kaid & Holtz-Bacha, 2006, p. 12}). These messages give the parties the chance to reach and persuade a heterogeneous and geographical dispersed audience, without losing control of the messages, unlike the news media with their routines of information selection and treatment.

On the other hand, a message is negative if it attacks or criticizes a campaign opponent, in particular when it targets his personal or leadership treats, integrity or competence, previous political performance, or its views about the campaign issues and proposals (^{Walter & Vliegenthart, 2010}). Even when these messages seem unambiguously beneficial to the attacker, since they hinder the credibility and eligibility of its opponent, they can be harmful for two reasons. First, they use resources that could be spent on self-praising, and second, the sponsor could be morally sanctioned by the voters because of the frequency and hostility of those attacks (also called “boomerang effect”). Since the decision to attack has several advantages and drawbacks, the question is what factors raise or lower the incentives of the candidates to emphasize the weaknesses of their opponents in terms of their ideas, failures, and personal mistakes, instead of self-praising their own personal or political strengths (^{Lau & Rovner, 2009}).

Some theoretical perspectives pose different sets of variables that mold the content of campaign messages (^{Esser & Strömbäck, 2012}; ^{Mancini & Swanson, 1996}; ^{Norris, 2000}), that include aspects of the political system (government, electoral and party systems), campaign regulations (its length, time allocation, censorship norms) and the level of professionalization (specialization, high budgets, aggressive consultants (^{Hill et al., 2015}), as well as some traits of political culture (voter turnout levels, party identification).

Nonetheless, the most empirically robust work about negative strategies on campaign communications (^{Walter, 2013}, ^2014a, ^2014b) stems from the rational choice theory of political behavior, inspired by the work of ^{Robert Downs (1957)}. Where several actors compete for the availability of the electoral market, that is, for the undecided voters -that change over time-, they act by a rational principle of maximizing the communication resources they have, to influence the voter’s choice. To that end, a possible strategy consists of self-praise of the candidate on behalf of their credibility, integrity, charisma, or policy program. The opposite strategy is to diminish the praise or support of the opponent by the undecided voter, by attacking him (^{Lau & Pomper, 2001}; ^{Walter, 2014b}; ^{Walter & van der Brug, 2013}).

Negative advertising does that. As a persuasion resource, it has several benefits: it is memorable and flashy, it coerces the opponents to reply to it -getting them out of their agenda- and get conspicuous media coverage, given the newsworthiness of political conflict (^{Geer, 2012}; ^{Ridout & Walter, 2013}; ^{Salmond, 2011}; ^{Sullivan & Sapir, 2012}). On the opposite, as we mentioned, a boomerang effect could ensue and decrease the support for the sponsor because of the disgust of the public about such messages (^{Skaperdas & Grofman, 1995}). In light of these facts, campaigns calculate the utility -in terms of costs and benefits- of going positive or negative, and how aggressive the latter should be. Nonetheless, this is pondered in a non-leveled field of competition, where contestants have asymmetric resources during a protracted campaign where many incentives to attack could arise. These conditions are operationalized in the literature in several predictive variables about the behavior of the campaigns, though we point out three of them in the Mexican case.

A first incentive to attack is the position of the candidates in the polls, linked to the availability of the electoral market. In political systems with three competitive parties (like the one we analyze), the frontrunners in the polls are the most attacked, since their competitors can extract more votes from them and they tend to attack less, fearing a boomerang effect. If the gap of the frontrunner is wide enough, he will probably not attack at all, but if the margin is less comfortable, he will not attack his weakest rival (^{Skaperdas & Grofman, 1995}). On the contrary, laggards tend to attack more often than frontrunners, and those on the third or fourth positions attack the frontrunner, but not each other (^{Salmond, 2011}). This behavior leads to conclude that the closest the races, the greater the utility of attacks, since the available market of undecided voters is so small, and the victory so close, that laggards take the risk to face a boomerang effect. That is why the literature agrees widely that the tighter the races, the more negative they become (^{Desposato, 2008}; ^{Papp & Patkós, 2017}; ^{Salmond, 2011}; ^{Walter, 2014a}; ^{Walter & van der Brug, 2013}).

Secondly, the asymmetries in communication resources constrain their strategies. Challengers tend to attack more often than incumbents, that praise the success of their policies to get them or their party reelected. Challengers have, on the contrary, more incentives to attack: they are prone to criticize the performance of the administration, emphasize an argument for change or replacement, and are willing to face the risk of a boomerang effect since they have less to lose (^{Lau & Rovner, 2009}).

Last, the ideology of the parties is relevant in at least three ways. First, is less likely that ideologically similar parties attack each other, so they do not alienate their same base, while parties of opposite ideologies have more incentives to attack each other since they naturally disagree on many issues (^{Geer, 2012}). Second, emergent or minority parties are more likely to attack established parties, as they are part of the status quo and can get more votes from them than their small party peers (^{Lau & Rovner, 2009}; ^{Salmond, 2011}; ^{Sullivan & Sapir, 2012}; ^{Walter, 2013}, ^2014b). Third, and particularly in post-authoritarian countries, previously hegemonic parties tend to attack based on their ample institutional experience and sophisticated campaign tactics (^{Sullivan & Sapir, 2012}).

Yet, most of the research and hypothesis discussed have been developed for the American broadcast system, with an almost nonexisting regulation about freedom of expression that has barely changed in recent decades (^{Kaid, 2006}). On the contrary, international comparisons as Holtz-Bacha’s (2017) and ^{Kaid’s (2006)} demonstrate that the traits of political advertising are closely related to the material and regulatory conditions of the media systems where they are broadcast. Free time allocation or purchase only, strict rules or open programming, length of the spots and total length of the campaign, the public or private broadcast system and censorship norms -be it moral, nationalist or political- and the quality of sanctions, exert an influence on the structure and content of the ads and the frequency of positive or negative spots (^{Kaid & Holtz-Bacha, 2006}).

Certainly, there are more factors or antecedent conditions to this phenomenon, but these are the more salient in a rational choice perspective of the behavior of candidates in campaigns. Before posing specific hypotheses, it is important to reflect on the weight and performance of these factors in the Mexican scenario.

Negative advertising in Mexico

Academic awareness about negative advertising in Mexico began in the 2006 presidential campaigns when this format was used in an unprecedented frequency and aggressiveness. Along with the very closed final tally (0.56%) it raised concerns among politicians and scholars. Though a minimal effect from the ads on voter intention was demonstrated in that race (^{Guerrero & Arellano, 2012}; ^{Moreno, 2004}; ^{Sánchez & Aceves, 2008}), content analyses differ on their findings, depending on the field where they were produced. Qualitative or impressionistic analyses claim that negative tactics were widespread, full of slander, and aimed to cause fear in voters (^{Castañeda & Coutiño, 2016}; ^{Chihu, 2011}; ^{Gutiérrez, 2007}). In contrast, quantitative research on this and further campaigns find the opposite, that the attacks, while certainly a component of Mexican campaigns, are less frequent than issue proposals and even negligible (^{Díaz & Alva, 2016}; ^{Echeverría & Juárez, 2011}; ^{Juarez, 2007}; ^{Juárez & Brambila, 2013}; ^{Juárez & Echeverría, 2009}).

In addition, the characteristics and limits of negative advertising were defined in the electoral reform of 2007-2008, that changed the sponsorship of the ads, from the parties to the State, and hardened the sanctions and criteria about slander or defamation in the spots, as a way to tackle the political parties’ outrage of the 2006 campaign. These measures were further reinforced in the 2014 reform. Thus, academic research began to ponder if those reforms had any effect on the traits and frequency of the ads. For ^{Castañeda and Coutiño (2016)}, the goal of the reforms to lower the frequency failed, since they found a very much aggressive midterm election in 2015. This conclusion is shared by ^{Lugo (2011)}, who thinks that the 2007 reform not only failed to reduce negative ads but increased them and made them subtler as a way to avoid official sanctions.

Beyond the antecedent condition of regulation, only the work of ^{Díaz & Alva (2016)} has considered other factors that can explain the variations in frequency and traits of negative advertising. They find that the ads get more negative if they are produced by the local campaigns rather than the national parties, if the candidates are challengers instead of incumbents, or if they are laggards in the polls (^{Díaz & Alva, 2016}). This paper extends those findings but adds a couple of procedures: it uses more robust statistical tests (multiple regression and bivariate logistic regression) and analyses more cases that come from the last four presidential elections. In this way, the factors discussed in the previous section can be operationalized as explanatory variables of the ads.

Method

A content analysis was conducted on the produced negative spots of 2000 (14), 2006 (16), 2012 (33) and 2018 (45), for a total of 108 pieces. The 2012 and 2018 were gathered from the National Electoral Institute, whereas the 2000 and 2006 ads were recovered from YouTube and other websites. We selected the ads upon the directional definition of negativity, that is, the messages that depict the rival candidate or party in an unfavorable light (^{Benoit & Sheafer, 2006}).

Though many codebooks are available in the international literature, we used the instruments developed and tested in Latin America, in terms of reliability and validity, and that are culturally closer to our case (^{Freidenberg & González Tule, 2009}; ^{García Beaudoux & D’Adamo, 2006}; ^{Pérez Dámazo, 2014}). From these codebooks, we selected the frequent categories that allow describing negativity, organized in the two channels used to utter an attack, the verbal and the visual one.

As for the former, we coded variables that expressed a higher or lower intensity of the attacks. Then, we codified whether they are direct, if they mention the opposite candidate or party, or indirect, that subtly make a nod about them. Also, if they use emotional appeals, that arouse feelings from viewers, or logical, that present facts or arguments to rationally persuade them. We also coded variables that operationalize the traits of the attacks. This includes the target of the attack, whether is the stand of the adversary about some issue, their biographic background, their previous performance in office, their personality traits, their partisan or ideological membership, or their campaign strategies. On the other hand, tactics were coded, be it humor, negative association with previous assertions, identification of the opponent with failed or controversial policies, prominent citizens or celebrities that criticize the opponent, or comparisons to the opponent.

Positive ads from every election were included in the sample too (N=287), in order to test the independent variables that predict the production of negative versus positive advertising. Thus, a single variable with two categories was used.

As for the second, visual channel, it was coded in the assumption that words, images, and sounds in a spot “interact with each other to heighten the main message and its emotional impact” (^{Richardson, 2001, p. 777}). Therefore, we coded the variables number of cuts, presence of music, types of shots (over the shoulder, long shot, medium shot, etc.), and overall color (black and white, color and sepia) according to the experimental variables summarized by ^{Juárez (2009)}, that have demonstrated cognitive effects in commercial advertising. To allow complex statistical tests we created an index with those variables, that operationalize the visual intensity of an ad. The index has a major value when an ad has more cuts, closer shots to the attacked candidate, has music, and is colored in an evocative way (either in black and white or sepia). We consider that those stimuli heighten the intensity of the message in the visual channel. Consequently, those variables, coded as nominal, were converted to interval ones.

As for the independent variables, we took the electoral reform (before and after 2007), party position (incumbent or challenger), the difference between the winner and the closest contender in the final tally -as a proxy for the race competitiveness-, as well as the political ideology of the sponsors, either parties or coalitions, based on their manifestoes or international memberships (as the leftist Centrist Democrat International, or the rightist International Socialist). Hence, we tagged the National Action Party (PAN, in Spanish) as right, the Revolutionary National Party (PRI) as center-left, and the Democratic Revolution Party (PRD) and the Movement of National Regeneration (MORENA) as left. Minority parties coalesced with the dominant ones, except in 2006 and 2012, when the Social Democratic Party and New Alliance, respectively, raced individually (with two ads each one).

To explore variable associations and their intensity, we used Chi-Square with Cramer’s V tests of hypothesis. In dichotomous nominal variables, we used a binomial logistic regression and for the index, we ran a multiple hierarchical regression. For the latter, the nominal variables were converted to dummy. Those tests were carried over the ads of the four elections, instead of on the ads from each of them, so the statistical outcomes could be more robust.

Pieces were coded by a single graduate student, though intercoder reliability was used, which was acceptable at 0.68 Cohen’s Kappa.

Given the theoretical and empirical insights from previous sections, some research questions and explanatory hypotheses are posed, all of them framed by a rational choice perspective.

RQ1. What is the effect of the electoral reform on the frequency and characteristics of negative advertising? The hypotheses follow as H1. The reform lowers the likelihood of negative ads being aired; H2. The reform lowers the likelihood of direct and emotional attacks; H3. The targets and tactics of the attacks are different after the reform; H4. The reform lowers the visual intensity of the attacks.

RQ2. What is the effect of the position of the party in power, as incumbent or challenger, in the frequency and characteristics of negative advertising? The hypotheses are H5. Incumbent parties are less likely to air negative ads than challengers; H6. Incumbent parties are less likely to air direct and emotional ads than challengers; H7. Incumbent parties point to different targets and tactics than challengers in their ads; and H8. The ads aired by incumbent parties are more likely to use more visually intense attacks than challengers’.

RQ3. Does the competitiveness of the campaign have an effect on the frequency and characteristics of negative advertising? The hypotheses are: H9. Competitiveness increases the likelihood of negative advertising being aired. H10. The competitiveness of the election makes it more likely to produce direct and emotional attacks; and H11. Competitiveness makes it more likely to air visually intense attacks.

As for the ideology variable, the fact that Mexican parties raced in coalitions makes it difficult to test for minority or ideologically akin parties, as some papers do. Therefore, we test the effect of the party’s ideology in the content of the ads, posing the exploratory question: Q4. What is the effect of the party’s ideology on the frequency and traits of the negative advertising?

Findings

We describe findings according to each of the explanatory variables used in the model (because of space limitations, descriptive tables, and multiple hierarchical regressions are not detailed in the paper). Electoral reform is a strong predictor of the changes in the content of the ads (Table 1). Type of attacks, X² (1)= 13.749, p=.002, the targets of the attacks, X² (6)=18.983, p=.004 and its tactics, X² (6)=14.252, p= 0.027, are strongly associated to that regulation (Cramer’s V= .362, .417, .363, .214, .260, respectively). Indirect attacks raise 38% after the reform, from 4% to 42%. A similar trend occurs with the targets of the ad since attacks to partisanship raises 10% and personal traits, 14%. Similarly, after the reform certain attack tactics, like humor, association to infamous characters, and comparisons to candidates, increase 10%, 4%, and 16%, respectively.

Moreover, the binary logistic regression (Table 2), shows that the reform raises the likelihood for attacks to be direct by 18.732 times (p=.008), though it does not increase the likelihood for them to be more or less emotional.

As for the visual variables, the reform is associated with the coloring of the ad X² (1)=5.009, p=0.025, and presence of music X² (1)=7.238, p=.007, though narrowly (Cramer’s V =.214 y .260, respectively). After the reform, the ads are flashier, as they increase the cuts from 2 to 7 on average, and include more music, since the ads that used music raised from 17% to 39%. Nevertheless, the reform does not predict the overall visual intensity of the spots.

On the other hand, political position, as incumbent or challenger, is strongly associated with the types of appeal, X² (2)=14.712, p=.001, targets of the ad X² (6)=32.973, p=.000 and tactics, X² (6)=15.312, p=.018, though not to the types of attack nor none of the visual variables coded. Thus, emotional appeals increase 27% from the incumbent to the challenger, who uses them in 95% of the spots; conversely, logical appeals are used by the challenger only in 5% of the pieces, while the incumbent uses them in 33% of their attacks. Challengers target their attacks to partisanship in 24% of the pieces, and campaign strategies 22% (just 2% and 0%, respectively in the case of the incumbent). Also, they emphasize the personal traits of their opponent (26%) and their previous record in office (52%) more often than the challenger (12% and 22% respectively). Challengers differentiate from incumbents by using negative associations to policy proposals (26% against 13% from the incumbent), while the latter tend to associate their opponents with infamous characters (17% vs 3% by challengers) and blaming them for a negative record in office (34% against 21%).

However, binary logistic regressions did not find statically significant predictions for this independent variable. That is the case, also, for the index of visual intensity.

As for the independent variable of competitiveness, the fact that is an interval variable does not allow us to run Chi Square associations. Nonetheless, binary regression demonstrates that competitiveness increases the likelihood of negative ads to be aired, by 1.046 times. Other variables did not show statistical significance, including the visual index.

Lastly, as for the party’s ideology, this is associated with the type of appeal X² (4)=11.482, p=.022, and the target of the attack X² (12)=46.31, p=.000. Parties from the right are more prone to logical appeals (27%) than emotional ones (which are overridingly used by other parties) and tend to use direct attacks more often (75%) than the parties from the right (55%).

As for the targets of the attack, each party emphasizes some of them. The left underscores opponents’ partisanship (40%) and personal traits (20%), the center-left the latter (39%), and their proposals (34%). The attacks of the right are based on the record in office of their opponent (54%), their stance on issues, and partisanship (14% each).

In fact, according to binomial regressions, the likelihood that the ads are negative if sponsored by a center-left party is 3.21 higher than when sponsored by a right party. It is an independent variable that has significant effects in the model, X² (5)= 37.927, p=.000, explains 13% of the variance (Nagelkerke R2) and classifies correctly the 72% of the cases, though only that dependent variable has statistically significant effects. No effects were found about the index of visual intensity.

Discussion and conclusions

As hypothesized (H2, H3) the electoral reform is a strong predictor that lessens the negativity of the ads, as the legislators intended. Since its passing, attacks are more indirect; their targets and tactics changed, with an emphasis on partisanship and personal traits. Reform is also linked to some indicators of visual intensity, since they have more music, have evocative colors, and are more dynamic. However, the reform does not predict that negative ads air more often than positive ones (H1) or that the visual variables get more intense (H4). According to the theory of campaign regulation (^{Kaid, 2006}; Holtz-Bacha, 2017), that predicts significant changes in the content of the ads if regulated -especially pertaining censorship norms- our findings support those of ^{Lugo (2011)} though contradict ^{Castañeda & Coutiño’s (2016)}, and ^{Juárez & Brambilla’s (2013)}. It seems that the reform decreased negativity in the ads, making them subtler, that is, indirect and visual, rather than verbal. Perhaps the parties were deterred by possible sanctions if their ads were overly aggressive, so they changed the intensity and tactics of the attacks, but not its frequency.

As for the position in power by the sponsor, be it incumbent or challenger, it led to significant statistically differences in regard to the target and tactics of the spots (H7), that are focused, in the former, on the partisanship of the opponent and the campaign strategies, emphasizing personal traits and opponents’ record in office. As we pose in the H6, the challenger uses more emotional appeals, though that variable could not predict any changes in the range of emotions. Nonetheless, this condition does not have an effect on the likelihood of airing negative advertising (H5), the utterance of direct attacks (H6), or the visual traits of the messages (H8). As a result, we found partial evidence that supports the hypothesis that challengers attack more and with different strategies than the incumbents, as is usual in the United States (^{Lau & Rovner, 2009}) and was found in Mexico (^{Díaz & Alva, 2016}).

This difference is probably due to the traits of the Mexican political system, where no reelection is possible and the extent of the federal administration is lengthier (6 years) than in the U.S. Both factors reduce the power asymmetries between incumbent and challenger parties, since the former is more likely to enter the race with a weakened political capital, and can not concentrate their resources on the same person.

The competitiveness of the campaign has a small effect in either verbal and visual variables, though is relevant that it increases 1.05 times the likelihood of negative advertising being aired (H9). Therefore, the findings of international literature are not supported (^{Desposato, 2008}; ^{Lau & Rovner, 2009}; ^{Papp & Patkós, 2017}; ^{Salmond, 2011}; ^{Walter, 2014}; ^{Walter & van der Brug, 2013}), though those works include more cases, that may make their tests more robust.

Last, party ideology is important in the type of appeal. The right is more straightforward and logical than the left in their attacks and targets the record in office of the opponent. It is striking, too, that the likelihood that a left-center party attack triples in comparison with the right. It seems that there are consistent patterns of strategies depending on the political ideology of the sponsor, as mentioned in the literature (^{Sullivan & Sapir, 2012}).

In sum, presidential candidates in the last four presidential races in Mexico have deployed certain negative strategies and tactics depending, overall, on the current regulation and the ideology of the parties. Competitiveness is not a variable that predicts changes in the traits of the ads, though it does raise the likelihood of the ads to be negative rather than positive. The position of parties in office does influence the content of the attacks, but not significantly. According to the data, going negative in Mexico depends mostly on external or preexisting factors, as the current regulation or party’s ideology, rather than the specific situations on a given campaign, be it the position in the polls or asymmetries of political resources.

More research is needed to thoroughly understand the antecedent conditions of negative advertising. A limit of our study is the scarcity of cases given the few races observed (only the four post-authoritarian ones), which in turn reduces the statistical robustness of our tests. Thus, we recommend widening the campaigns observed to include midterm races or local ones, raising the number of cases. Likewise, it is advisable to conduct a comparative design between countries, to test the influence of national systemic conditions. In the face of toxic campaign environments, where negativity seems to be at the forefront, it is needed more research about what raises it and eventually how to tame it.