Introduction

The federal electoral process 2017-2018 was characterized not only for had been the largest in Mexico’s history for the number of popularly elected positions that were elected,¹ but also for had been the most violent. From the beginning of the process, on September 8, 2017, until the end of the campaigns on June 27, 2018, 48 pre-candidates and candidates were assassinated throughout Mexico.² Undoubtedly, the issue of electoral violence, in particular the murders of candidates for the 2018 federal and local elections, was one of the topics that attracted more media attention during this election. The Organization of American States (OAS), the European Union delegation in Mexico, together with the embassies of Norway and Switzerland, expressed their concern about violence and intimidation during said electoral process. Various political and social actors also expressed their concern about the murder of candidates. In the current context of widespread criminal violence, it is worthwhile asking whether we can still affirm that “On the shiny surface of Mexico’s democracy, things by and large seem fine” (^{Schedler 2014: 10}), or whether we are witnessing the beginning of a “societal [criminal/] subversion of democracy”.

What was behind this unprecedented wave of political assassinations? Does the violence respond to incentives of electoral competition, or can it be attributed to criminal organizations? Our argument is that the murders of candidates observed in Mexico in 2017 and 2018 are not the result of fierce political competition between political parties, because unlike other electoral systems, in Mexico access to power and political competition has largely followed the institutional path. Instead, the murder of candidates can be attributed to the activities of criminal organizations in the municipalities in which the murders occurred. Our statistical evidence suggests that in some states the candidates were among civilian casualties in the midst of criminal violence that the country is experiencing; however, in the states of Puebla and Guerrero, it is possible to affirm that the murdered candidates were targeted by criminal organizations. The phenomenon of electoral violence also seems to be a local phenomenon, since most of the victims were competing for positions at the municipal level. In a democracy in process of consolidation it is important to determine if the deaths of candidates had their origin in political processes, that is, whether they are the product of a process of institutional weakening in which political actors from enclaves of sub-national authoritarianism resort to murder as a strategy to perpetuate themselves in power; or, whether the murders are the product of contexts of criminal violence.

Therefore, this article has two objectives. First, distinguishing whether the wave of murdered candidates is rooted in political motivations based on electoral competition, or if they are the result of strategies undertaken by criminal organizations. During the electoral process, two possible hypotheses were discussed as potential causes of the ongoing murders. The first related to electoral competition. In other parts of the world it has been observed that electoral violence arises particularly in situations in which elections offer a real possibility of changing existing power relations, and it is precisely in closed elections in which politicians have incentives to make use of violence. The second hypothesis that was put forward was that criminal organizations could have used political assassinations as a strategy to politically capture state officials, and thus obtain a comparative advantage over other criminal organizations. Second, after identifying that the assassination of candidates is related to the presence and operation of criminal organizations, we set out to determine whether the murdered candidates were a specific target of the criminal organizations, or if they were victims of the environment of generalized violence, that is rampant in many areas of the country. To do this, we use an exact binomial test to test the hypothesis that candidates at the state level are more likely to be killed than the general population.

This article is organized as follows. First, we review the literature on crime and electoral violence, and the incentives that both criminal organizations and political parties have to make use of it. Then, after a thorough review of the press and the literature, we present some theoretical expectations about the causes of the murders of candidates in Mexico, that is, whether they are due to electoral competition or if they are related to the activities of criminal organizations. Next, we present a quantitative analysis of the municipalities that are located in states that suffered at least one murder of pre-candidates or candidates (710 municipalities excluding Oaxaca), in order to test both narratives and to identify the political and social patterns that are correlated with the murders of candidates. In the conclusions, we present the research findings and the implications for the quality of democracy.

Electoral violence: electoral competition or criminal organizations?

Electoral violence has two main meanings. Firstly, it can be understood as a subset of activities within a larger political conflict, most of the time as part of a trajectory of ethnic violence within divided societies (Fjelde and Höglund, 2014; ^{Wilkinson, 2004}). Secondly, it can be related with the intentional use of force as a strategy to influence electoral results (^{Hafner-Burton et al., 2014}; ^{Höglund, 2009}). Electoral violence is usually employed by two main actors; first, by governments and political parties —to remain in power or to harm their contenders— (^{Hafner-Burton et al., 2014}; ^{Höglund, 2009}); and second, by criminal groups that seek to intimidate their opponents or to support allied politicians and thus obtain protection for their illegal activities (^{Albarracín, 2018}; ^{García-Sánchez, 2016}; ^{Ley, 2018}; ^{Ponce, 2019}; ^{Trelles and Carreras, 2012}). Depending on who is the perpetrator, electoral violence will have different logics because it serves different objectives.

First, we will examine electoral violence exerted by governments or by political parties. Incumbent parties can resort to various electoral manipulation strategies, perhaps the most extreme of which is the use of violence. Electoral violence arises particularly in situations where elections offer a genuine possibility of changing existing power relations. Therefore, political parties facing closed elections will have greater incentives to promote violence as a mean of gaining power (^{Höglund, 2009}, ^{Strauss and Taylor, 2009}). In addition, there are other factors that interact with electoral competition and that promote the use of violence; for example, an incumbent party has incentives to use pre-electoral violence if it anticipates an unfavorable electoral result and if there are also weak institutional controls on the executive (^{Hafner-Burton et al., 2014}). For example, in Africa, incumbent parties have been associated with the use of political violence, especially in systems with poorly consolidated democracies (^{Onapajo, 2014}; ^{Strauss and Taylor, 2009}) and with weak institutional controls (^{Onapajo, 2014}).

In addition to electoral competition, there is also a link between electoral violence and electoral rules. Majority electoral systems are prone to generate violence in places where ethnic groups have been systematically excluded from power and where there are significant economic inequalities. In contrast, the proportional representation rules make it likely that even ethnic minority candidates will commit to respecting the rules of democratic competition, as these electoral rules guarantee them a minimum of representation to remain in the electoral arena until the next election (Fjelde and Höglund, 2014). A minimum of political representation reduces the incentives of political parties to resort to electoral violence.

For their part, criminal organizations also have incentives to get involved in electoral processes. Not only in Mexico, but in other parts of the world such as Italy (^{Alesina et al., 2016}; ^{Dal Bó et al., 2006}; ^{Daniele and Dipoppa, 2017}), Brazil (^{Albarracín, 2018}) and Colombia (^{García-Sánchez, 2016}) there are examples of how criminal groups get involved in political affairs. It is usually assumed that criminal organizations act mainly motivated by economic interests; for example, to defend and control criminal markets (^{Osorio, 2015}). However, they also act motivated for political ends, or in reaction to political variables.

The literature recognizes at least four reasons why criminal organizations attack sitting politicians or candidates. First, traffickers around the world usually avoid open confrontation, and prefer less visible techniques (such as corruption) to minimize the impacts of police action and law enforcement; but when the state undertakes repressive measures indiscriminately, drug traffickers respond by violently attacking state agents (^{Lessing, 2015}).

Second, turf wars between different criminal organizations induce cartels to change the balance of the electoral game in order to capture state agents, and thereby gaining an advantage to control the market. Politicians can become the target of criminal violence when they decide not to cooperate with criminal organizations and instead they fight illicit markets (^{Albarracín, 2018}). Consequently, criminal groups use violence to instill fear among voters and negatively affect voter turnout (^{Ley, 2018}; ^{Trelles and Carreras, 2012}), often with the purpose of reducing the chances of victory for political opponents in areas over which they exercise territorial control (^{García-Sánchez, 2016}). In extreme cases they try to eliminate electoral competition through threats or assassinations of rival politicians (^{Albarracín, 2018}; ^{Daniele y Dipoppa, 2017}; ^{Ponce, 2019}). It is worth mentioning that criminalized electoral politics, in which there is an intentional use of force by criminal groups to influence politics, is predominantly a local phenomenon. Criminalized electoral politics may occur at the national or regional level, but a weak rule of law and limited accountability to civil society are required to make it attractive to criminal groups (^{Albarracín, 2018}).

Third, armed groups can attack sub-national authorities in an attempt to extract economic resources from local governments (^{Cubides, 2005}), and use them to strengthen their military presence (^{Chacon, 2018}). This rent capture strategy is activated only if the amount of resources they can extract exceeds the costs of corrupting or coercing the authorities (^{Dal Bó et al., 2006}). Finally, drug trafficking organizations can reap multiple benefits if they manage to capture sub-national elections: greater facilities for the transport of illegal drugs, fewer obstacles to money laundering, intelligence information, achieving protection from the local police, and even getting support from the police to combat rival organizations (^{Ponce, 2019}). For this reason, electoral cycles, particularly electoral campaigns, stimulate violence since it is more costly for criminal organizations to capture sitting politicians, who in theory have greater police protection, than candidates seeking power (^{Ponce et al., 2019}). Consequently, Mexican cartels take advantage of the opportunities offered by local electoral cycles, when some politicians are vulnerable because of the lack the protection from central government authorities, and decide to attack mayors or candidates to gain control over local governments and thus are able to establish subnational criminal governance regimes (^{Trejo and Ley, 2019}).

Candidate’s murders in Mexico

During the 2017-2018 electoral process, various actors expressed their opinions about the possible causes of the wave of candidates murdered in Mexico. In a press conference offered on May 27, 2018, the secretary of the Interior, Alfonso Navarrete, assured that the candidate murders were not linked to ideological reasons or to a fierce electoral competition, and “that among the causes of these homicides were disputes between criminal organizations, and issues of a personal and family nature” (^{Salgado, 2018}). For his part, Rubén Salazar, director of Etellekt, a consultancy firm that developed a detailed count of attacks against candidates in 2017-2018, stated that a possible explanation for the murders was that “democracy is no longer the tool to settle differences in political competition” and that violence is being used “as the most effective resource to obtain public office and to get rid of opposition candidates” (^{EFE, 2018}). Which of the two narratives best fits the causes of candidate killings? Should they be attributed to the conditions of electoral competition, or to the activities of criminal organizations?

A first approach to the causes of the murders of candidates and pre-candidates in Mexico is to make a general description of the main characteristics of the victims. As indicated in footnote number two, the data we use on candidate killings comes from reviewing newspaper notes.³ The count we carried out showed that between September 8, 2017 — the beginning of the federal electoral process — and June 27, 2018 — the end of the electoral campaigns — 48 politicians were assassinated; 20 were already officially registered candidates and 28 were pre-candidates.⁴ If we compare the 2018 figure with other federal and local electoral processes of the last ten years in Mexico, we will observe a dramatic growth in the number of murdered candidates. In previous electoral processes (federal and local) the number of murdered candidates ranged from two to eight (Figure 1).

Source:^{Integralia Consultores (2015).}

FIGURE 1 Candidates assassinated in Mexico: recent evolution 

Out of the 48 assassinated politicians, 26 aspired to a municipal presidency, eight to be local legislators, 10 to a local council and two more to be a federal legislator. These figures, in accordance with the literature (^{Albarracín, 2018}), show that the victims were mostly candidates for positions at the local level. On the other hand, the figures of these murders show that, from the total number of murdered candidates, 41 were men and seven were women. This makes us suppose that the murders are not entirely random. If they were, we would observe a similar proportion of murders between men and women. This, because in Mexico it has been possible to postulate the same proportion men and women as candidates as a result of the adoption of gender quotas and, more recently, by the adoption of gender parity (^{Freidenberg and Alva, 2017}).

Regarding the political affiliation of the victims, 12 belonged to the PRI, 11 to the PRD, six to Morena and five to the PAN (Figure 2). Since practically all the political parties that nominated candidates for popularly elected positions were victims of political violence, it seems that political orientation is not a variable that can be associated with the occurrence of murders.

Source: Own elaboration from journalistic notes.

FIGURE 2 Candidates assassinated by political party 

The state that registered the more candidates assassinated, according to the information gathered, was Guerrero, with 12; Michoacan, Oaxaca and Puebla follow with five murders each, and the State of Mexico and Jalisco with four murders (Figure 3). This suggests that the murders of candidates occurred mostly in the states most affected by criminal violence.

Source: Own elaboration from journalistic notes.

FIGURE 3 Candidates assassinated by state 

Data and methods

To test the above hypotheses, we carried out a quantitative analysis at the municipal level in all the states in which murders of candidates occurred in the 2017-2018 electoral process. This represents 909 municipalities in 12 states of the country (see Figure 3). The municipality was chosen as the unit of analysis since most of the assassinated candidates aspired to positions at the local level (mayors, councilmen, and local deputies); only one candidate for a federal office (federal deputy) was killed.

To carry out the analysis, we included two alternative measures of the dependent variable. First, we have a dichotomous variable that identifies whether or not the murder of at least one candidate or pre-candidate occurred in the municipality in question. This variable takes the value of one if at least one murder occurred, and zero otherwise. The second measure of our dependent variable is a count of the number of candidates killed by municipality. In some municipalities, such as Chilapa, up to three murders of candidates occurred in the past electoral process (Table 1 A, in the appendix, shows the descriptive statistics of the variables used).

The measures of the independent variables are three. First, the number of criminal organizations that operate in each municipality to test the hypothesis about the impact of organized crime organizations on the murders of candidates (^{Coscia and Ríos, 2012}). Second, we include the homicide rate per 100 000 inhabitants per municipality, reported in the first half of 2018, just before election day, as a proxy for criminal violence.⁶ This variable is continuous and its values range from 0 to 131 homicides per 100 000 inhabitants. Third, our electoral competition measure is instrumentalized as the subtraction of the percentage of votes obtained by the first place minus the percentage obtained by the second in the 2018 municipal elections.⁷

The controls included in the analysis are variables that the literature has identified as possible predictors of electoral violence, in particular for the murders of candidates. The first control has to do with the functioning of the institutions in charge of imparting justice and the presence of a rule of law. An efficient legal system orders social relations, but its absence or the “brown areas” with less state presence lead to problems of governability (^{O’Donnell, 2004}) and insecurity (^{Aguirre and Herrera, 2013}). To measure the impact of the institutional strength and presence of the state, we included the number of officials of the Public Prosecutor’s Office at the state level.⁸

It has also been argued that alternation in the governor’s office can produce higher levels of violence, as the rotation of parties in the state government erodes informal protection networks that had facilitated the operation of organized crime organizations under unified governments ( ^{Trejo and Ley, 2018}). To test this effect, we include a dummy that identifies the states in which there was an alternation before 2018. Likewise, it has been argued that political violence is more common at the local level and in smaller communities (^{Albarracín, 2018}; ^{Daniele y Dipoppa, 2017}), therefore we include a variable that measures the percentage of the population that lives in populations with less than 2500 inhabitants.⁹ Additionally, we include the incidence of robberies at the municipal level, measured with the number of robberies with violence of passers-by on public roads.¹⁰ This variable was included because in the press review we observed that in some cases the murders of the candidates were the product of robbery with violence, so it could be assumed that a higher incidence of robberies at the municipal level is positively related to the murder of candidates. Finally, we include the Municipal Human Development Index to control for structural factors of violence such as quality of life.

Victims of general violence or the target of criminal organizations?

In the previous section, we established that the murder of candidates is related with the presence and operation of criminal organizations in their municipalities. However, this statistical relationship does not allow us to determine whether the murdered candidates were a specific target of criminal organizations or if they were victims of the environment of violence that exists in many areas of Mexico. The homicide rate nationwide in the first half of 2018 was 29.27 per 100 000 inhabitants, while the rate of murdered candidates and pre-candidates nationwide was 55.39 per 100 000 candidates. These figures indicate, at first glance, that during the 2017-2018 electoral process, more candidates died than ordinary citizens. This leads us to raise the following proposition:

P1. The probability of being murdered among the candidate subpopulation is greater than among the general population.

Following ^{Bartman (2018)}, who analyzes the murder of journalists in Mexico, we used an exact binomial test to identify whether candidates at the state level are more likely to be killed than the general population. The exact binomial test evaluates whether there is a significant deviation from the binomial distribution. It is appropriate to use it when there are two discrete results for the same categorical variable, for example two different results for each state. The null hypothesis is that the ratio of successes equals a particular ratio. In this case, the test calculates the probability of observing k (or more than k) candidates killed in a state with a probability p of homicides for the general population, given that the probability of being killed would be the same for both candidates as for the general population (Equation 1). If the test is significant it indicates that the probability of candidates murdered differs from that of the general population.

Where, n = number of candidates by state, p = probability of homicides of the population in the state, and k = number of candidates killed in each state. This estimation requires the number of candidates killed by state,¹⁵ the total number of candidates who participated in the electoral process in the state (Table A3, in the appendix),¹⁶ the number of homicides by state,¹⁷ and the general population by state.¹⁸

When we compare the homicide rates per 100 000 inhabitants with the rate of candidates assassinated, it turns out that the rate of murdered candidates is higher than that of the general population in the states of Quintana Roo, Colima, Guerrero, Puebla, Jalisco and Coahuila (Figure 4). However, in the states of Quintana Roo and Colima, there was only one candidate assassinated, so its apparent high rate of candidate homicides may be due to the fact that this single homicide is being weighted by a very small population of candidates. In the rest of the states the rate of murdered candidates is equal to or less than the rate of the general population.

Source: Own elaboration.

FIGURE 4 Comparison of probabilities of homicides of candidates and the general population 

To determine if these differences are significant we applied the exact binomial test. Table 3 shows the p-values of the exact binomial test for the 12 states in which assassinations of candidates and pre-candidates occurred in the 2017-2018 electoral process.¹⁹ The homicide rate differs significantly in only two states: Puebla and Guerrero. Therefore, in these two states, proposition 1 is confirmed: the probability of being murdered among the candidate subpopulation is greater than among the general population. The states with the highest homicide levels are not per se the states that experience the most homicides against candidates in the Mexican elections. From the evidence presented in this section, we can conclude that the presence and operation of criminal organizations can produce the murder of candidates as a collateral damage, that is, that they do not necessarily focus on assassinating them; but in the states of Puebla and Guerrero, there is statistical evidence suggesting that the high murder rate of candidates was the product of strategies in which criminal organizations targeted candidates.

Discussion

Electoral violence, understood as the intentional use of force as a strategy to influence electoral results, can be used by governments and political parties (^{Hafner-Burton et al., 2014}), as well as by criminal groups (^{Albarracín, 2018}; ^{García-Sánchez, 2016}; ^{Ley, 2018}; ^{Ponce, 2019}; ^{Trejo and Ley, 2019}; ^{Trelles and Carreras, 2012}). The murders of the pre-candidates and candidates who participated in the 2017-2018 electoral process have multiple causes. Based on a statistical analysis of the municipalities located in the states affected by political violence, we find that the murders of candidates in the past electoral process are related to the presence (number) of criminal organizations and with the homicide rate in the municipalities where the murders occurred, as well as the same rate for those municipalities where the candidates competed. Considered together, the evidence gathered from reviewing of the press and the evidence from the statistical analysis suggest that organized crime activities are one of the most important variables behind the killings of candidates in the electoral process. Unlike other regions of the world, electoral competition does not appear to be a direct cause of the murder of candidates in Mexico.

Having established that the killings of candidates are related to the activities of criminal organizations, we proceeded to assess whether the killings are the product of a strategy in which the candidates are the specific target of criminal organizations, or whether their deaths may be attributed to the general context of violence in the country. In two states of the country (Puebla and Guerrero) we find that the murder rate for candidates differs significantly from that of the general population, and therefore it is feasible to assume that in those states the murdered candidates were the target of criminal organizations.

Additionally, the statistical analysis suggests that there is a negative relationship between the number of public prosecutors operating in each state —a proxy for state capacity— and the number of candidates murdered in each municipality. This indicates that a greater presence of state security agencies, particularly those in charge of initiating and directing criminal investigations on behalf of the state, may be an element that dissuades members of criminal organizations from carrying out acts of violence against the candidates. Similarly, ^{Trejo and Ley (2019: 2)} also conclude that criminal organizations particularly attack and threaten politicians who “are politically and militarily unprotected by the central authorities.”

These findings allow us to argue that, as the costs of exercising electoral violence increase, with more agents in charge of activating and exercising criminal justice on the streets, the less likely that assassinations of candidates will occur. Conversely, when the costs of exercising electoral violence decrease, with more criminal organizations operating in a given municipality and a higher homicide rate, it is more likely that political assassinations will increase. If we consider both findings together, we could argue that electoral violence is primarily the product of material incentives to compete among criminal organizations and of a weak rule of law.

What are the implications of these findings for the quality of democracy? What is the role that criminal organizations will play in the short and medium term in Mexican democracy? Although it has been argued that criminal organizations have neither the means nor the intention to appropriate the state or its territory (^{Lessing, 2015}), nor to model electoral governance institutions at their whim (^{Schedler, 2014}), there is no doubt that the recent wave of murdered candidates has damaged the electoral arena. The security of local candidates in many states of the country does not seem to be guaranteed. If the fact that running for popularly elected positions at the local level in certain regions of the country represents a risk to the lives of the candidates: Who will run for office? Will the number and quality of candidates decrease? ^{Ponce (2019)} has found that the number of candidates decreases significantly in the most violent municipalities. Future investigations could also determine whether in the states most affected by the murder of candidates there is a decrease in the years of experience of the candidates, or in other indicators of their quality. Additionally, and thinking about designing public policies aimed at reducing the incidence of this problem, pilot programs could be implemented during electoral periods in which the presence of public prosecutors is increased in the areas most affected by violence against candidates. Although we observed a significant increase in the number of assassinated candidates in the 2017-2018 electoral process, we are still in time to contain its expansion and harmful effects.