English (pdf)
Article in xml format
Article references
Automatic translation
Send this article by e-mail
Cited by SciELO 
Cited by Google 
Similars in
SciELO 
Similars in Google 
Permalink
Introduction
This article addresses the reasons for the failure of the anti-corruption referendum in Colombia. Colombia is a country with a significant corruption problem. Transparency International’s 2017 Corruption Perceptions Index places Colombia 96th out of 180 countries (Transparency International 2018). Although more than 99% of the voters who came to the polls voted in favor of the proposals, participation was very low. This raises an important question. If corruption is an important issue, why were citizens not mobilized to vote in the referendum? In this article, I examine the factors that influenced the outcome of the referendum at the municipal level.
This article contributes to the current global debate on direct democracy. In the context of Brexit, the issue of direct democracy and citizens’ decision-making has become a frequent area of interest in research (Becker, Fetzer and Novy 2017; Goodwin and Heath 2016). Moreover, researchers have also explored direct democracy in Colombia, whether the peace agreement referendum in 2016 (Dávalos et al. 2018; Liendo and Braithwaite 2018; Matanock and García-Sánchez 2017; Matanock and Garbiras-Díaz 2018; Rincón Morera 2018; Tellez 2018; Mejía-Cáceres 2018), recall elections (Welp and Milanese 2018; Eberhardt 2018), or other referendums (Dietz 2018). The factors influencing outcomes and participation in referendums have long interested researchers (Altman 2011; LeDuc 2007; Svensson 2002; Franklin, Van Der Eijk and Marsh 1995; Renwick 2017). This article, therefore, builds on this research by adding new findings on the results of the 2018 referendum in Colombia. In addition, this article also contributes to the debate on the role of corruption in electoral behavior (McCann and Domínguez 1998; Stockemer, LaMontagne and Scruggs 2013; Sundström and Stockemer 2015; Karahan, Coats and Shughart 2006; Escaleras, Calcagno and Shughart 2012). It is particularly interesting to examine the role of corruption and whether it mobilizes or demobilizes citizens in voting on anti-corruption measures. I use OLS (ordinary least squares) regression models to examine the effects of individual variables in 1,101 Colombian municipalities.
I have divided this article into four parts. In the first part, I briefly describe the context of the Colombian referendum. In the second part, which is theoretical in nature, I formulate hypotheses from existing research. I first examine corruption and its impact on electoral behavior, and then review which variables are significant in explaining the outcome of referendums, focusing mainly on political variables and the role of ideology. The third part discusses methodology. I describe my dependent, independent, and control variables, the origin of the data, and the reason for including these variables in the regression models. In the last part of the article, I evaluate my findings and interpret the results of the regression models.
1. The background of the Colombian referendum
Colombia has long had a corruption problem. Its citizens are, of course, concerned about this situation. According to 2014 AmericasBarometer data, 60% of Colombians believe that corruption is very widespread amongst public officials and 24% believe that it is somewhat widespread (Latin American Public Opinion Project 2015). According to supporters of the anti-corruption referendum, its proposals would constitute one of the first steps in reducing corruption. The referendum (la consulta popular) is, according to Colombian law(1), an opportunity for citizens to express their opinions on important issues. The spokesperson for the referendum was Claudia López, who started to collect signatures on January 24, 2017 along with Angélica Lozano and other members of the promotion committee. The organizers collected 4,226,682 signatures, which they brought to the National Registry Office or Registraduría Nacional del Estado Civil (Registraduría) within six months. The Registraduría acknowledged 3,092,238 as valid signatures. This exceeded the 5% threshold necessary in order for a referendum to be held. Thirty-five thousand volunteers were involved in the signature collection process. The Senate of the Republic of Colombia approved the referendum on June 5, 2018. The motion passed with 84 Senators voting in support and none voting against. President Juan Manuel Santos signed decree 1028/2018 on June 18, 2018, completing the process of approving the referendum. The referendum took place on August 26, 2018. There were seven questions in total. Citizens voted on each one separately. The referendum concerned the introduction of measures(2)to curb corruption (Registraduría 2018a; Andrés Sánchez 2017; “Colombia hopes” 2018).
One possible reason the Senate approved the referendum without opposition, with even Álvaro Uribe voting for it, is that the senators voted before the second round of the presidential election. Any opposition to the anti-corruption referendum could have resulted in the loss of crucial votes in the second round. The organizers of the referendum included more than just members of civil society. López and Lozano were Senators for the Green Alliance. The government party, the Democratic Center, was not unanimous. Newly elected President Iván Duque did not mention the referendum in his inaugural speech. However, he avoided calling for a boycott of the referendum and ultimately said he would vote in it (“Consulta anticorrupción divide” 2018; “Colombia hopes” 2018; “Cinco razones” 2018).
There are two critical points to note about the referendum. Firstly, its opponents pointed out that it could be considered ineffectual because the proposed measures would not be effective in fighting corruption or were already enshrined in law. Secondly, opponents also criticized the cost of organizing the referendum: 300 million Colombian pesos (“Consulta anticorrupción divide” 2018; “Consulta anticorrupción valdrá” 2018; “Colombia hopes” 2018).
It was very problematic for the Democratic Center to support the referendum. One of the main problems was incarnated by Claudia López. This former vice-presidential candidate was perhaps the most visible leader of the referendum, although, in the last few weeks before the vote, the referendum became a national topic supported by various politicians and members of civil society. When the collection of signatures began, it was López who was the face of the organization of the referendum. There is clear animosity between López and Uribe, which continues to this day. For example, López has called Uribe a murderer (“Consulta anticorrupción divide” 2018). From a political point of view, it was thus very difficult for Uribe to support a referendum organized by his strong opponents. The success of the referendum would have been perceived as a success for his rivals, which could have complicated regional elections in 2019.
Theories and hypotheses
The relationship between corruption and political participation
In this article, I use the same definition of corruption as (Walczak 2018, 256): “an abuse of power, influences, professional position for one’s individual interests and goals.” Therefore, I view corruption in broad terms. Under this definition, corrupt practices in the form of abusing power for private interests and benefits can nonetheless be legal under the penal code. For example, legal corruption (Kaufmann and Vicente 2011) may exist in a relationship between well-connected firms and political elites through legal lobbying or legal contributions to political parties. Moreover, these practices flourish in an environment of low transparency. Proponents of the anti-corruption referendum argued for changes to the law in order to reduce potential corrupt practices and to raise transparency.
Many researchers have examined the impact of corruption on political participation and turnout. There is no clear answer to the question of whether corruption mobilizes or demobilizes citizens. Some argue that corruption has a negative impact on participation and discourages citizens from voting (McCann and Domínguez 1998; Stockemer, LaMontagne and Scruggs 2013; Sundström and Stockemer 2015; Chong et al. 2015; Simpser 2012). (Similarly, Miles 2015) analyzes 35 advanced democracies using surveys and aggregated data. He finds that citizens are more likely to vote in countries where institutions govern more fairly, enforce the rule of law, and control corruption.
(Anderson and Tverdova 2003) show that the perception of corruption is an essential determinant for evaluating political systems in 16 European countries. They also analyze trust in civil servants and find that greater corruption leads to less trust in officials. However, the negative impact of corruption is weaker among citizens who support the incumbent government. (Warren 2004) claims that corruption shows a deficit of democracy. The problems that corruption poses for democracy can be quite significant. Corruption violates several of the principles that democratic societies stand for and which citizens count on. (Seligson 2002) studied corruption in four Latin American countries and states that corruption lowers confidence in the political system and even damages relations between people and reduces interpersonal trust. Other researchers have confirmed the negative impact of corruption on perceptions of democratic institutions (Linde and Erlingsson 2013; Wagner, Schneider and Halla 2009; Mishler and Rose 2001). (Bauhr and Grimes 2014) investigate whether exposure to corrupt practices causes resignation or indignation. They find that when transparency is increased in countries with high levels of corruption, there is more resignation than indignation. (Dahlberg and Solevid 2016) carried out a multilevel analysis that combined data from the Comparative Study of Electoral Systems at the individual and national levels in 26 countries. They find that corruption reduces turnout, but only in countries with a low to medium level of corruption. The effect of corruption disappears in countries with a high level of corruption.
In contrast, another group of researchers argues that corruption increases voter turnout. However, this group is in the minority. These researchers argue that politicians attempt to gain reelection through pork barrel spending. Public officials take advantage of their positions and mobilize citizens to go to the polls by essentially bribing them. This behavior leads to a higher chance of reelection. Public officials spend more money and effort on their campaigns and electorate when the value of office is considered very high (Karahan, Coats and Shughart 2006; Escaleras, Calcagno and Shughart 2012). (Inman and Andrews 2015) report, based on a field experiment, that Senegalese citizens who perceive more corruption are more likely to vote.
However, studies that explain corruption as a reason for higher turnout and mobilization are not usually applicable to the Colombian referendum. This is because, in Colombia, no office holders were directly affected by the outcome of the referendum. The referendum was not about specific politicians but about proposals introducing anti-corruption measures.
There are several studies on corruption at the municipal level. (Stockemer and Calca 2013) find that corruption is a strong mobilization factor in Portuguese municipalities. They point out that corruption and perceptions of corruption may differ between national and sub-national levels. A country with high levels of corruption and low turnout can actually have high turnout in the most corrupt municipalities (Stockemer and Calca 2013, 536). Comparative research cannot verify this at the national level. On the other hand, (Giommoni 2017) finds that instances of corruption in Italian municipalities decrease turnout. (Costas-Pérez 2014) shows that corruption at the level of Spanish municipalities lowers voter turnout. However, corruption scandals discourage only those citizens who are independent and without strong political attachments. Corruption does not affect the core supporters of the incumbent or opposition.
(Kostadinova 2009) argues that corruption is a significant factor affecting turnout. Nevertheless, the relationship between voter turnout and corruption is complex. In cases where voters expect their choice to cause real change, they will mobilize and vote for another candidate to remove corrupt politicians. The mobilization effect ceases to exist when citizens do not expect change and believe that their vote will make no difference. (Haveric, Ronchi and Cabeza 2018), using data from the World Values Survey, find that the impact of corruption is not the same for all citizens. While corruption reduces citizen participation, this does not apply to state employees who vote in a highly corrupt environment. (Carreras and Vera 2018) find, using the AmericasBarometer survey, that high corruption demobilizes citizens in Colombia. Demobilization also occurs when corrupt politicians provide public works to their constituencies, casting doubt on the theory that voters can be bought.
In this article, I investigate corruption indirectly, given the difficulty of obtaining municipal data on corruption. It is not clear how corruption can be measured even at the municipal level with a sufficient data sample. The existence of legal corruption, as mentioned above, makes the situation still more problematic (Kaufmann and Vicente 2011). Corruption in the form of diverse networks of influences, links, and connections (Walczak 2018) is almost impossible to quantify. Thus, even if data on politicians convicted of bribery were available, it would not offer a full picture of all corrupt practices. I therefore examine the transparency of municipalities. Although corruption may of course be present in transparent municipalities, the evidence in the literature suggests that the more transparent the municipality, the less space for corruption. Kaufmann and Bellver (2005) demonstrate a strong positive correlation between political and institutional transparency and lower corruption. (Wehner and De Renzio 2013) find that increased fiscal transparency is associated with a reduced level of corruption. (Ellis and Fender 2006) study the levels and growth rates of output and find that lower corruption depends on the transparency of the fiscal system. (Matheson and Kwon 2003, 15) consider inadequate compliance with accounting and reporting rules and weak internal and external controls to be contributing factors in potential corruption.
It is precisely such gaps in the governance of municipalities that open government indices seek to solve. One step toward reducing corruption is raising transparency within government institutions (Tanzi 1998, 122-123). (Martins et al. 2018) demonstrate a strong correlation between levels of corruption in different countries and the United Nations E-Government Development Index, which measures the role of e-government in the political system. Electronically available data increases the transparency of government institutions. (Jiménez and Albalate 2018) examine the relationship between transparency and the occurrence of corruption in Spanish municipalities. They find that the higher transparency, the lower the probability of corruption. A lower degree of transparency increases the risk of potential corruption, and the authors state that transparency is a good proxy for the likelihood of corruption.
One of the main motivations for launching open government data initiatives is to reduce corruption. State authorities introducing these indices expect to increase transparency (Attard et al. 2015).
The Office of the Inspector General of Colombia, or Procuraduría General de la Nación (Procuraduría), created the Open Government Index or Index Gobierno Abierto (IGA) in 2010 with the aim of reducing corruption and improving public administration. In creating the IGA, the Procuraduría was inspired by Donald (Cressey’s theory of fraud Cressey 1953), which states that there must be three factors present (the fraud triangle) in order for fraud to occur: pressure, opportunity, and rationalization (that is, the personal justification of actions). The Procuraduría was particularly interested in reducing the opportunity for fraud through monitoring and through promoting compliance with anti-corruption rules (Procuraduría General de la Nación 2011).
My first hypothesis is therefore the following, based on the research material on the negative effects of corruption, including research conducted specifically in Colombia at the individual level (Carreras and Vera 2018):
H1: The more transparent the municipality, the higher the percentage of the municipal electorate that voted in favor of proposals in the anti-corruption referendum.
Turnout, ideology, and outcomes of referendums
One question to be addressed is why the referendum voter turnout was low even though more than 99% of participating citizens supported the proposals. (Altman 2011, 23) points out that in some situations when a quorum is present, it is better for referendum opponents to use demobilization strategies than to mobilize voters to vote against referendum proposals, especially when it comes to highly emotional issues. Demobilization efforts will affect both citizens who agree with the referendum proposals and citizens who disagree with them. Furthermore, encouraging participation could lead to the necessary referendum quorum being reached, causing the referendum to be valid. Therefore, in cases where turnout is expected to be low, opponents might prefer voters to be absent. This strategy could explain why, in Colombia (which has traditionally low turnouts), politicians did not attempt to mobilize voters to vote against the referendum proposals. (Schuck and de Vreese 2009) examine the effects of mobilization, using content analysis of newspaper and TV news about the 2005 Dutch EU Constitution referendum. They find that the efforts of the referendum organizers also mobilized skeptics.
There is a significant amount of research on quorums. (Maniquet and Morelli 2015) argue that it is better to use the approval quorum than the participation quorum, as it obviates the shortcomings of the participation quorum. This is the minimum threshold that must be exceeded for a proposal to pass. In the case of an approval quorum, none of the parties need to use the demobilization strategy, and this maximizes the number of citizens who can go to the polls. (Aguiar-Conraria, Magalhães and Vanberg 2016) claim that quorum rules reduce voter turnout and promote election boycotts. This effect is much greater in the case of a participation quorum than an approval quorum. (Hizen and Shinmyo 2011) argue that with a high enough threshold, the status quo is more likely to be maintained if voters decide not to vote than if they were to vote. A number of studies have concluded that a participation quorum has a negative effect on turnout (Herrera and Mattozzi 2010; Aguiar-Conraria and Magalhães 2010) and have examined the quorum and the obstacles it presents (Uleri 2002).
There is also a substantial amount of research on the relationship between referendums and turnout. (Butler and Ranney 1994, 17) observe that turnout is usually lower in referendums than in general elections. However, despite this, (LeDuc 2007, 27) points out that voter turnout can be very high if the referendum places a crucial issue before voters. There is therefore a great difference between referendums. Turnout is directly related to voters’ interest in the issue and the intensity of the campaign (Kriesi 2007). (Renwick 2017) studies 21 countries with experience of direct democracy and finds that there is indeed higher fluctuation in voter turnout in referendums than in national elections. This confirms LeDuc’s (2007) observation. Campaign intensity is a significant factor in explaining turnout or the outcome of a referendum. If voters are exposed to more information during a campaign, they are more likely to vote in a referendum. Citizens also feel they understand more about an issue in the case of a referendum (Hobolt 2007).
The existing literature offers many explanations of how citizens decide which way to vote in a referendum. One explanation is that citizens have their own feelings and opinions on the matter in question. For example, there have been referendums on issues regarding the European institutions (Siune and Svensson 1993; Garry, Marsh and Sinnott 2005; Svensson 2002). These concern fundamental issues for citizens, who vote according to their beliefs and attitudes. On the other hand, when the issue involved is not very important to citizens, or they do not understand it, they get advice from others. Politicians and political parties try to inform their supporters about the matter being addressed in a referendum. In this context, the referendum may also become a second-order election (Reif and Schmitt 1980), and possibly be used to punish the governing parties. In such a case, citizens’ primary motivation is to show their dissatisfaction to political leaders (Franklin, Marsh and Wlezien 1994; Franklin, Van Der Eijk and Marsh 1995). A poorly informed voter might also take cues from political representatives as information shortcuts for decision-making. This decision could be similar to a well-informed voter’s (Lupia 1994; Bowler and Donovan 1998). In these cases, the recommendations of politicians are crucial.
It is vital to examine political support and corruption together. (Charron and Bågenholm 2016) explain the importance of ideology in the context of corruption in 24 European countries. They utilize hierarchical models based on data at the individual and national level and find that the more citizens fall toward an extreme of the political spectrum, the more likely they are to ignore corruption allegations and continue supporting their party. There are two recent studies about the Colombian peace agreement referendum in 2016. (Liendo and Braithwaite 2018) use individual data from public opinion research and find that citizens’ decisions were driven more by political preferences than real conflict experiences. Similarly, (Dávalos et al. 2018) use hierarchical models and study the relationship between the result of the referendum and, among other variables, political support for the Democratic Center. They find that support for the party strongly affected support for the peace agreement. Other researchers have also examined this question (Matanock and García-Sánchez 2017; Matanock and Garbiras-Díaz 2018; Rincón Morera 2018; Tellez 2018; Mejía-Cáceres 2018).
This article’s second hypothesis focuses on whether political affiliations also affected the 2018 referendum. Fajardo ran for president and López ran for vice-president and was also the leading spokesperson for the referendum. Therefore, the second hypothesis is as follows:
H2: The higher the support for Fajardo, the higher the support for the referendum proposals.
Uribe, the founder of the Democratic Center, was probably the most visible opponent of the referendum. Duque ran for president for the Democratic Center. Therefore, the third hypothesis is as follows:
H3: The higher the support for Duque, the lower the support for the referendum proposals.
3. Methodology
Dependent variable
The dependent variable is the percentage of the municipal electorate that voted in favor of the seven questions in the referendum. The voter could vote differently on each of the questions, but over 99% of citizens who came to the polls agreed with all seven questions. Therefore, differences in voting among the participating citizens were minimal. However, for the sake of accuracy, I added up all “yes” votes among the municipal electorate for all the questions and divided this amount by the number of proposals (seven). I therefore calculated the average number of votes in favor of the proposals. I used this number as a dividend and the municipal electorate as a divisor. In this way, I arrived at the share of the municipal electorate that voted in favor of the measures in the referendum.(3) The data was obtained from the Registraduría (Registraduría 2018a).
Independent variables
There are two primary types of independent variables in the research. The first builds on a theory of corruption and transparency. As outlined in the previous section, the Registraduría created the IGA to monitor public management at the municipal level. The IGA should serve as a preventive strategy in the fight against corruption; the Registraduría evaluates municipalities in its final annual report. The methodology for calculating the IGA is quite complex and considers a number of variables that offer information about transparency, compliance with regulations, and other conditions related to fighting corruption (for more details, see the Registraduría, [2018b]). The Registraduría created the first index in 2012 and the last in 2016. For this reason, and to capture the previous management of a municipality, I used the IGA variable as a five-year mean of the indices in the municipality from 2012 to 2016. The municipal elections took place in Colombia in October 2015; the five-year mean covers both the current municipal leadership and the previous administration. Therefore, it offers more information about the long-term context of municipalities.
Political independent variables involve the percentage of votes in the first round of the 2018 presidential election. I chose the presidential election because presidential elections are more important than parliamentary elections in Colombia, given that it is a country with a presidential system. The main spokesperson for the referendum, López, ran for the position of vice-president, and Fajardo ran for president. My reason for selecting the first round is that Fajardo came in third in the first round and was therefore eliminated before the second round. Apart from the votes for Fajardo, the other political variables are the vote percentages for the current president, Duque, as well as Gustavo Petro and Germán Vargas Lleras. Even though Petro and Vargas Lleras fall outside the scope of the hypotheses, I included them in separate models as political variables because they received enough votes to be relevant. I did not include the other candidates because they received around 2% or less of the votes. The data was obtained from the Registraduría (2018c).
Control variables
In general, turnout in a referendum is affected by the same sociodemographic factors as turnout in a general election (LeDuc 2007; Hobolt 2007; Neijens et al. 2007). Income, education, and living conditions affect citizens’ participation in elections. I controlled for these factors and used the Multidimensional Poverty Index (MPI) to comprehensively measure poverty in Colombia. The advantage of the MPI is that it accounts for not only income but also education, health, and living standards. Unfortunately, the latest data for Colombia is from the 2005 census. The data from the 2018 census was not yet available. Using 2005 data for the MPI should not be an obstacle because I used data from 2012 to 2016 to calculate the 5 year mean of the IGA, and a high correlation can be expected(4) between the 2005 and 2018 MPI data.
It is important to control for voter turnout in a general election. For example, (Neijens et al. 2007) use turnout from previous elections as a control variable in their Amsterdam referendum research and demonstrate its significance. (Geys 2006) demonstrates in his meta-analysis of voter turnout that previous turnout is statistically significant for explaining turnout in subsequent general elections. Therefore, I used turnout (as a percentage) from the first round of the presidential election in 2018 as a control variable.
(Powell 1984) examines participation, stability, and violence in democracies in his work. He finds that countries with high voter turnout have on average the least amount of deaths by violence (1984, 26). (Fornos, Power and Garand 2004) find that political violence in Latin America reduces turnout in legislative and presidential elections. (García-Sanchez 2007) finds that violence has a negative effect on political participation in Colombia. I used the number of murders per 100,000(5) inhabitants to express the amount of violence in a municipality.
I controlled for another two variables. The first variable was population size and the second, municipalities’ level of urbanization. All control variables were based on data from the Panel Municipal provided by the Centro de Estudios sobre Desarrollo Económico-La Universidad de los Andes (CEDE-Uniandes). The source of the panel data is the DANE, except for homicide rates, which were obtained from the Ministry of National Defense (CEDE 2018).
Models
I tested the hypotheses in five(6) models. In the first four models, the control variable of previous turnout was not included. This is because the variable itself accounts for 56.9%(7) of the variance in the dependent variable. Individual presidential candidates were analyzed in separate models to avoid multicollinearity(8) and misinterpretation. The fifth model included previous election turnout as well as the vote percentages for Duque and Fajardo. I chose these candidates for the fifth model because they had almost no(9) correlation and, more importantly, because Fajardo’s vice-presidential candidate, López, was one of the main organizers of the referendum. Since Uribe, a prominent member of the Democratic Center, was against the referendum, I used his party’s candidate, Duque, as another variable. The highest variance inflation factor (VIF) was 3.17 in the first four models. The mean of VIFs was 1.51 for model one, 1.92 for model two, 1.55 for model three, and 1.52 for model four. In model five, the highest VIF was 3.95, and the average value was 1.99. Therefore, none of these models had high multicollinearity. Descriptive statistics for all the variables are presented in Table 1.
4. Results
Before interpreting the regression models testing the hypotheses, I report the results of the regression where the dependent variable is turnout in the first round of the 2018 presidential elections in Table 2. I call this Model 0. The model includes the IGA and control variables. In this sense, the results of the referendum can be generalized to turnout in various types of elections in Colombia.
Table 1 Descriptive statistics of dependent and independent variables
| Variable | Observations | Minimum | Maximum | Mean | Standard Deviation |
|---|---|---|---|---|---|
| Electorate that voted for proposals | 1101 | 3.67 | 54.03 | 26.98 | 9.25 |
| IGA | 1101 | 38.64 | 89.95 | 67.07 | 7.88 |
| Iván Duque | 1101 | 3.61 | 87.99 | 48.01 | 18.36 |
| Sergio Fajardo | 1101 | 0.37 | 49.80 | 13.42 | 9.71 |
| Gustavo Petro | 1101 | 1.07 | 88.65 | 23.73 | 18.31 |
| Germán Vargas Lleras | 1101 | 0.92 | 63.68 | 10.11 | 8.79 |
| MPI | 1101 | 14.27 | 100.00 | 69.22 | 16.23 |
| Homicide rate | 1101 | 0.00 | 281.37 | 23.96 | 30.46 |
| Population (logged) | 1101 | 2.98 | 6.91 | 4.17 | 0.50 |
| Urbanization | 1101 | 1.71 | 99.91 | 44.94 | 24.22 |
| Turnout in the presidential election | 1101 | 17.53 | 75.80 | 50.28 | 9.47 |
Source: Author's calculations
Table 2 shows that the IGA and control variables apart from urbanization are statistically significant. Turnout was lower in municipalities that are larger, less transparent, have more people living in poverty, and have more murders. The range between the municipality with the highest IGA and the lowest is 51 points. Therefore, this model predicts that the difference in voter turnout between the best-rated municipality and the worst would be almost 9% if all other variables remained constant. It might be expected that these independent variables would have a similar effect on turnout in the referendum. Below, I explore the significance of this variable when controlling for the presidential election.
Table 2 OLS regression - 2018 presidential election
| c | Model 0 |
|---|---|
| Dependent variable: Turnout in the presidential election | |
| IGA | 0.173*** (0.032) |
| MPI | -0.344*** (0.018) |
| Homicide rate | -0.066*** (0.007) |
| Population (logged) | -2.594*** (0.490) |
| Urbanization | -0.017 (0.011) |
| Constant | 75.667*** (3.519) |
| N | 1101 |
| R2 | 0.482 |
Note: Standard errors in parentheses, *p < 0.05, **p < 0.01, ***p <0 .001 Source: Author's calculations
Table 3 OLS regression - 2018 anti-corruption referendum
| IV | Model 1 | Model 2 | Model 3 | Model 4 | Model 5 |
|---|---|---|---|---|---|
| Dependent variable: Percentage of municipal electorate that voted for proposals | |||||
| IGA | 0.187*** (0.030) | 0.106*** (0.031) | 0.186*** (0.030) | 0.106*** (0.031) | 0.057* (0.024) |
| Iván Duque | -0.100*** (0.011) | -0.131*** (0.009) | |||
| Sergio Fajardo | 0.177*** (0.032) | 0.189*** (0.025) | |||
| Gustavo Petro | 0.118*** (0.012) | ||||
| Germán Vargas Lleras | -0.125*** (0.025) | ||||
| MPI | -0.387*** (0.016) | -0.304*** (0.021) | -0.421*** (0.017) | -0.36*** (0.017) | -0.116*** (0.018) |
| Homicide rate | -0.064*** (0.006) | -0.065*** (0.007) | -0.063*** (0.006) | -0.063*** (0.007) | -0.032*** (0.005) |
| Population (logged) | -2.520*** (0.479) | -1.261*** (0.463) | -2.757*** (0.477) | -1.222** (0.464) | -1.564*** (0.375) |
| Urbanization | -0.086*** (0.010) | -0.081*** (0.011) | -0.081*** (0.010) | -0.078*** (0.011) | -0.071*** (0.008) |
| Turnout in the presidential election | 0.570*** (0.022) | ||||
| Constant | 61.919*** (3.398) | 48.951*** (3.420) | 57.483*** (3.254) | 56.180*** (3.379) | 16.721*** (3.158) |
| N | 1101 | 1101 | 1101 | 1101 | 1101 |
| R2 | 0.535 | 0.516 | 0.544 | 0.514 | 0.718 |
Note: Standard errors in parentheses, *p < 0.05, **p < 0.01, ***p < 0.001 Source: Author's calculations
Table 3 shows the individual models used to test the hypotheses, with the dependent variable being the percentage of the municipal electorate that voted in favor of the seven questions in the referendum. In the first four models, turnout in the presidential election was not a control variable. These four models explain between 51 and 54% of the variance of the dependent variable. At the same time, in all four models, presidential candidates are a statistically significant variable. However, while the percentage of votes for Duque and Vargas Lleras is negatively associated with the dependent variable, the percentage of votes for Fajardo and Petro is positively associated with it. These political variables are statistically significant in predicting the percentage of the municipal electorate that voted for the referendum proposals. The four models confirm the second hypothesis. Moreover, the first four models have a statistically significant five-year mean IGA variable. This variable affects results in the expected direction. The more transparent the municipality, the higher the support for the questions in the referendum. The first hypothesis is therefore confirmed. Furthermore, all the control variables are statistically significant in the first four models. Lower turnout in the referendum occurred in municipalities with a higher percentage of the population living in poverty. Murder rates had a negative effect on referendum participation. The size and urbanization of a municipality are negatively associated with the dependent variable. All control variables, except for urbanization, are statistically significant and in the same direction in these four models as in Model 0, where the dependent variable is turnout in the presidential election.
In the fifth model, I controlled for turnout in the presidential election. This model shows the strong positive effect on the dependent variable of turnout in the presidential election. This model explains 72% of the variance of the dependent variable. For every 1% increase in presidential turnout, the percentage of the municipal electorate supporting the referendum proposals increases by 0.57%. The other control variables remain statistically significant and are in the same direction as in the previous four models. However, the coefficient of the MPI variable decreases by a factor of approximately three. The fifth model shows that even controlling for turnout in the presidential election, turnout in the referendum in poorer municipalities is lower. The IGA coefficient also decreases in this case, but is still statistically significant. It appears that citizens living in non-transparent municipalities did not mobilize in this referendum but rather demobilized. Support for presidential candidate Fajardo in municipalities in the first round of the presidential election was on average 13.4% and at most 49.8%. In terms of the difference between his average and best result, the model predicts nearly 7% more support for the referendum proposals among the municipal electorate, when all other variables are controlled for. Duque is the opposite. Support for him in municipalities was on average 48% and at most 88% in the first round of the presidential election. In terms of the difference between his average and best result, the fifth model predicts over 5% less support for the referendum proposals among the municipal electorate, when all other variables are controlled for.
Assuming that transparency is a good proxy for corruption (Jiménez and Albalate 2018; Kaufmann and Bellver 2005), my results are in agreement with the research that has found that corruption has a negative impact on participation (McCann and Domínguez 1998; Stockemer, LaMontagne and Scruggs 2013; Sundström and Stockemer 2015; Chong et al. 2015; Simpser 2012). My research also confirmed the significance of political support in municipalities. It was possible to predict the municipal outcome of the referendum to a significant degree based on the support given to different candidates in the 2018 presidential election. These findings accord with previous research on political influence in the 2016 peace agreement referendum (Liendo and Braithwaite 2018; Dávalos et al. 2018).
Conclusion
I examined the relationship between transparency in Colombian municipalities and the results of the 2018 anti-corruption referendum. I also studied the outcome of the referendum in connection with the results of the 2018 presidential elections. The fact that the referendum was to a great degree organized by the Green Alliance, and particularly by López, had a significant effect on the subsequent results of the referendum. The citizens in Colombian municipalities where there was higher support for Fajardo were more likely to participate and vote for proposals in the referendum. The opposite was true in the case of Duque, because Uribe, the leader of the Democratic Center, was one of the primary opponents of the referendum. Therefore, political affiliations played an essential role in influencing citizens’ decision making.
This article contributes to the literature on whether voters in referendums vote primarily based on their own opinions or, instead, according to their party’s position. In the case of the Colombian anti-corruption referendum, party positions had a significant effect on citizens. What was unique about this referendum was that López’s support for the referendum and Uribe’s opposition to it greatly influenced the outcome, despite corruption being a major concern for Colombians. Nearly 12 million citizens decided to vote for the referendum proposals, but the referendum failed because it did not achieve the required quorum. The problem lies with Colombia’s traditionally low turnout, which is a relative exception in Latin America and is probably the result of non-compulsory voting. When a country uses the quorum system and generally has low turnout, the right strategy for opponents of a referendum is to boycott it with the aim that quorum not be achieved. In Colombia, this strategy meant that despite 99% acceptance of the referendum proposals amongst participating citizens, there was insufficient turnout for the referendum to be considered valid.
It is not clear from the current literature whether corruption mobilizes or demobilizes citizens to participate politically. This article explored that relationship, and the result is evident in the case of the referendum discussed. Colombian municipalities that are less transparent suffered from lower turnout in the presidential election and the referendum. Even after controlling for turnout in the previous presidential election, transparency remained statistically significant. The more transparent a municipality, the more support there was for the referendum proposals. This finding contributes to the contemporary literature on the influence of corruption and transparency on political mobilization. It is unique in that it concerns not just a general election but a referendum aimed at curbing corruption.
Further research could be directed at studying municipalities’ transparency. A lower level of participation in political decision-making in less transparent municipalities would constitute a problem for democracy, as there would be less pressure to increase transparency. Research should also be expanded beyond Colombia to other Latin American countries where corruption is a serious problem and where there is compulsory voting.
