Abstract
This article reports evidence concerning both the political manipulation in the regional allocation of the largest conditional cash transfer program in Colombia, and the political rewards that the incumbent national government obtained in the presidential elections by providing this program. To do this, we use a comprehensive data set for Colombian municipalities during the expansion period of the program between 2005 and 2009. Our evidence shows that, although the incumbent government followed the preestablished criteria for program allocation across municipalities, this manipulated the allocation of subsidies by targeting intensively swing municipalities and less intensively loyal municipalities. Our findings also show that, even after controlling for the effect of the political manipulation of the program on the incumbent vote share, the incumbent national government was politically rewarded by the electorate for providing subsidies through the program. Taken together, our results suggest that voters rewarded the incumbent governing coalition not only for redirecting the subsidies of the program but also for the allocation of subsidies in itself.
1 Introduction
During the last decades, several developing countries have implemented antipoverty programs in the form of conditional cash transfer (CCT) payments to lower-income families. There is an increasing interest in the literature in studying whether CCT programs have been used by politicians for electoral purposes (Fried 2012), and whether politicians have received political rewards for allocating these type of programs (Conover et al. 2019; De la O 2013; Galiani et al. 2017; Labonne 2013; Linos 2013; Manacorda et al. 2011; Rodríguez-Chamussy 2015; Zucco 2013). The aim of this article is to contribute to the corresponding discussion in the literature by providing empirical evidence for the largest CCT program in Colombia (called Familias en Acción, hereafter FA).
The Colombian’s CCT program is directly administrated by the national government. This defines the number of beneficiaries in each municipality by following both family eligibility criteria and criteria to determine the number of families covered by the program in each municipality. The program targets only poor families with children that are identified in all municipalities of the country through an economic status index. Broadly speaking, criteria to determine the number of families covered by the program in each municipality must take into account each municipality’s institutional capacity vis-à-vis the implementation of the program (in terms of health and education services). Furthermore, the program gives priority funding to those municipalities with the largest number of eligible families.
In this context, our aim is to study first whether the incumbent national government politically manipulated the allocation of the FA program across municipalities or not. Having identified the political motives behind the allocation of the program across regions (if any), and taking these motives into account, we investigate the existence of political rewards resulting from the allocation of the FA program. By doing this, we are able to analyze whether these electoral rewards (if any) were exclusively related to the political manipulation of the program or to the allocation of the subsidies in itself.
The unit of observation used in the analysis is the municipality. We concentrate on the 2006 and 2010 presidential elections in Colombia and the expansion of the FA program during this period. The main variables of interest are the FA beneficiary rate—defined as the number of beneficiary families (x 1,000 population), and the incumbent national government share of votes. Additionally, we focus on the role played by the concentration of swing versus loyal voters in determining both the allocation of the FA program across municipalities and voting behavior.
Therefore, in order to typify the political motives behind allocation of the CCT programs across municipalities, we utilize “loyal voter models” (Cox and McCubbins 1986), and “swing voter models” (Dixit and Londregan 1995, 1996; Lindbeck and Weibull 1987). The main difference between these categories of models has to do with the incumbent’s incentives to target specific segments of voters. We build historical measures of the respective concentrations of swing voters and loyal voters in each municipality. We use these measures to study the political motives behind allocation of the FA program across municipalities.
To identify the effect of the allocation of the FA program on the incumbent’s vote share—in the absence of suitable instrumental variables, we consider a large set of variables in our estimates and carry out additional tests to show the robustness of our results. First, we directly include in these estimates our measures of concentrations of swing voters and loyal voters. If there were political motives behind the allocation of the program and these motives are not taken into account when estimating the effect of the allocation of subsidies on voting behavior, then the estimated effect is likely to be biased. By the same token, considering these political incentives as part of the identification strategy helps to mitigate this problem. Second, we exploit in our estimates the panel structure of our data set so as to remove unobservable time–invariant municipal characteristics that might correlate with other possible strategies used by the incumbent in the allocation of the FA program. Third, we introduce in our econometric models a large set of control variables to avoid biases resulting from the omission of relevant variables. Finally, we conduct placebo tests to provide additional confidence for our results.
Our evidence shows that, although the incumbent national government followed the preestablished criteria for program allocation across municipalities, this manipulated the allocation of the FA’s subsidies by targeting intensively swing municipalities and less intensively loyal municipalities. An increment of one standard deviation in the concentration of a municipality’s swing voters implies an increment of 6.5 points in the FA beneficiary rate (13% of the mean). We claim that this allocation strategy is consistent with the political record of former president Uribe, the leader of the incumbent national government.
Furthermore, we find that the incumbent governing coalition was politically rewarded by the electorate for providing subsidies through the FA program, even after controlling for the effect of the political manipulation of the program on the incumbent vote share. An increment of one standard deviation in the FA rate in a municipality positively affected the incumbent governing coalition’s vote share in 2.94 percentage points (4.2% of the mean). Furthermore, consistent with our findings on the allocation of the program, we find that the incumbent governing coalition became more successful in capturing votes in historically swing municipalities and less successful in historically loyal municipalities. Taken together, these results suggest that voters rewarded the incumbent governing coalition not only for redirecting the subsidies of the program strategically but also for the allocation of subsidies in itself.
As far as we know, only Fried (2012) has analyzed the political motives behind the geographical allocation of CCT programs by using data for Bolsa Familia in Brazil. Interestingly, he finds no evidence for political manipulation related to the allocation of this program. Regarding political rewards, several recent studies have analyzed the relationship between CCT programs and voting behavior. Manacorda et al. (2011) find that beneficiary households of PANES in Uruguay were more likely to favor the incumbent who implemented the program. Zucco (2013) identifies a causal relationship between the allocation of the Brazilian CCT program (Bolsa Familia) and Lula’s vote share for presidential elections. De la O (2013) finds a positive effect of the allocation of Progresa in Mexico on the incumbent’s vote share for the 2000 presidential elections. Considering a CCT program in the Philippines, Labonne (2013) finds that in a competitive political environment, incumbent vote share was higher in municipalities with considerable coverage. Galiani et al. (2017) find a positive effect of a CCT program on the incumbent party share of votes during the 2013 presidential election in Honduras. Finally, using variation in the proportion of beneficiaries across voting booths within a polling station, Conover et al. (2019) find a positive effect of the allocation of the FA program on the incumbent governing coalition in the 2010 presidential elections in Colombia.
Our article contributes to this literature in two ways. First, unlike previous studies, it provides evidence of political manipulation in the geographical allocation of CCT programs. Second, by studying both the political motives behind the allocation of the FA program and its political effects on voting behavior, we are able to show that voters rewarded the incumbent national government not only for redirecting the subsidies of the program strategically but also for the allocation of subsidies in itself.
The rest of the article is organized as follows. First, we provide a description of the country’s political background and the FA antipoverty program. Second, we present our theoretical framework. Third, we describe the data set and present our results. Finally, we discuss these results and conclude.
2 Political Background
Colombia’s population is around 51 million inhabitants, with 40.3% of the population living below the poverty line in 2009 (49.7% in 2002).1 In Colombia, poverty exhibits a high degree of regional variability. For instance, in Choco (the poorest state in the country, with a population of 454,030), 68.3% of the population lives below the poverty line, whereas in Bogota (the capital city, with a population of 6,840,116), only 18.3% of the population are in the same situation. The county also has one of the worst income distributions in Latin America (a Gini index of 0.557 in 2009).2 This implies both that antipoverty programs are politically demanded by a large fraction of the population, and that direct income cash transfers represent a very important fraction of household incomes.
Previous to the Colombian’s 1991 Constitution, the country had a bi-party political system. The two hegemonic parties where the Partido Conservador and the Partido Liberal. With the aim of increasing the political participation of different sectors, the 1991 Constitution allowed for the formation of new political parties. As a consequence, several political parties were formed during this decade. The proliferation of parties during this decade was seen as a perverse phenomenon for the well behavior of the Colombian democracy and, at the end of the 1990s, a new reform was implemented to reduce the number of political parties and to improve their internal organization. Although this reform considerably reduced the number of parties, there were some national and regional parties emerging and disappearing from the political arena during the first decade of 2000. Interestingly, these new parties either preserved an ideological link with one of the two traditional parties or formed a political coalition with one of them. Moreover, many of the politicians who moved from one party to another have been active in politics for far longer. These facts help us to identify what we call the incumbent governing coalition in our period analysis and to properly define our swing and loyal measures.
The Colombian presidential electoral system is a majority one with a two-round system (i.e., if no candidate obtains more than 50% of the votes in the first round, there is a second round featuring the two candidates obtaining the largest number of votes). Presidential elections run every 4 years. Former president Andres Pastrana, an independent candidate supported by one of the main traditional parties in the country, Conservador Party, was in office when the FA program was implemented in 2001.
In 2002, the independent candidate Alvaro Uribe, with the support of a just-created political party called Primero Colombia, was elected in the first round as Colombia’s president for the period 2002–2006 (with 53% of the total votes). Notably, one of the main traditional parties in the country (Partido Conservador) decided to extend support to Uribe’s campaign. This support was accepted by Uribe, who ran 2002 elections representing Primero Colombia but with the support of the Partido Conservador. His political platform was mainly based on security issues and the fight against Colombian guerillas. Although his social policy proposals were concentrated on expanding coverage and quality improvements in education and social security services, he never included the FA program in his political program as an explicit tool for obtaining these objectives (see DNP, 2002).
A constitutional amendment in 2004 allowed incumbent president Uribe to run in the 2006 presidential election (Colombian Constitution used to prohibit presidential reelection). He was reelected that year in the first round (with 62% of total votes) for the period 2006–2010. Notably, as part of his political platform during this election campaign, Uribe introduced the FA program as the main mechanism for reducing poverty in the country, and proposed extending the program to include 1.5 million families (DNP, 2006). During his second term, Uribe governed with a political coalition made up of two relatively new parties, Partido de la U and Cambio Radical, together with the traditional party Partido Conservador. During his second term, the FA program was extended from 665 to 1,097 municipalities, and the number of beneficiary families grew by 278% (it went from 588,105 to 2,228,443 families).
Although a group of political parties and citizens tried to promote Uribe’s reelection a second time, it was declared unconstitutional, and the incumbent president was not allowed to run in the 2010 presidential elections. Each of the parties belonging to the governing coalition decided to run in the first-round presidential elections that year under a different candidate. No candidate was elected in the first round, but candidate Santos (of Partido de la U), with the direct support of president Uribe, was chosen to run in the second round. This time, the three parties belonging to Uribe’s governing coalition decided to support candidate Santos, and he was elected president with 69% of votes. Anecdotally, once in office, Santos began peace negotiations with the largest guerrilla group in the country, a fact that caused the breakdown of his relations with Uribe.
The above description allows us to identify what we call in our analysis the incumbent governing coalition, and, more specifically, the incumbent’s vote share in each electoral year. This vote share accounts for Uribe’s share of votes in the 2006 presidential elections. Since in the 2010 presidential election there were two rounds, we consider two measures of the incumbent’s vote share for this year. First, we consider the sum of the vote shares in the first round of the 2010 presidential elections of the three parties (Partido de la U, Cambio Radical, and Partido Conservador) belonging to the 2006–2010 governing coalition. Second, we consider Santos’s vote share in the second round of the 2010 presidential elections. Moreover, we use historical votes for these parties and the Conservador party to build our measures of the concentration of loyal voters and swing voters in each municipality. We come back to these measures in Section 5.
3 The FA Program
The FA antipoverty program was created by Colombia’s central government in 2001 (prior to Uribe’s first term in office), with the aim of alleviating and reducing poverty. Through an agency called Social Action Office (Oficina de Acción Social, nowadays called Social Prosperity Department—Departamento para la Prosperidad Social), the central government directly administers the program. Therefore,it is the incumbent national government, using the preestablished eligibility and allocation criteria of the program (see below), who decides the number of families to be covered in each municipality.
The FA program offers two types of subsidies. The first is a nutrition subsidy for families with children under 7 years of age. This subsidy allocates between 9 and 22 U.S. dollars per month to each family (according to regional location), regardless of the number of children in the family, though conditional on the family’s participation in health programs. The second subsidy is an educational subsidy to children 7 to 18 years of age, which allocates to each child between 7 and 52 U.S. dollars per month (according to regional location) conditional on school enrollment and certified attendance. These amounts have remained relatively constant (in real pesos) over time.
During our study period, there were both family eligibility criteria and criteria to determine the number of families covered by the program in each municipality. Only those families with children classified as SISBEN 1 in a municipality were eligible for the program. SISBEN is an index used to classify families according to their economic status. SISBEN 1 is the level assigned to those families suffering the worst conditions.3 This index is used to target different social programs in the country.
Criteria to determine the number of families covered by the program in each municipality exhibited some changes during this period. The program had two waves between 2001 and 2009 that coincided with each presidential period. The first wave of the FA program (the implementation wave, 2001–2005) was designed to provide direct conditional cash assistance to the poorest families in the country living in municipalities that were not departmental capitals, with populations of less than 100,000 persons, with at least one banking agency (through which to make the cash transfer payments, via the formal financial system), and with the institutional capacity to provide sufficient health and education services (in order to absorb the additional demand generated by the program). This institutional capacity and the amount of SISBEN 1 families in each municipality were also considered to determine the number of beneficiary families in each municipality.
Among the municipalities that satisfied the requirements mentioned above during this first wave (620 approximately), 270 municipalities were non–randomly chosen to implement the program in 2001. By 2002, the rest of these 620 municipalities were covered by the program. This first wave occurred before Uribe’s first term in office. During his first term, approximately 50 new municipalities were added to the program between 2003 and 2005 via presidential instruction. This reflects, as noted earlier, that the FA program was not a priority for the Uribe’s government during his first term in office.
During the second wave of the FA program (the expansion wave, 2006–2009, Uribe’s second term in office), the population, the banking agency and the nondepartmental capital constraints were eliminated. Nevertheless, the institutional capacity criteria (i.e., the capacity to provide health and education services), and the priority funding to those municipalities with the largest number of SISBEN 1 families were conserved to determine the number of beneficiary families in each municipality. As a result of these changes, the program was expanded to almost all of the municipalities in the country. This expansion of the FA program encompassed not only direct conditional cash assistance to the poorest families in the country, but also to people displaced due to internal conflict, and native communities. In 2009, 87% of resources were allocated to cash assistance to the poorest people. Our study concentrates on this subprogram, and does not take into account the subsidies allocated to displaced and native communities. Furthermore, we concentrate our analysis in this second wave of expansion.4
FA antipoverty program—Number of beneficiary families and percentage of municipalities with coverage 2001–2009. Source: Social Action Office. Own computations.
FA antipoverty program—Number of beneficiary families and percentage of municipalities with coverage 2001–2009. Source: Social Action Office. Own computations.
Although the central government and municipal mayors worked jointly at implementing the program in each municipality, as said before, it was the central government that decided the number of families to be subsidized in each municipality. Mayors collected the socioeconomic information of those families who claimed to be beneficiaries in their municipalities and submitted this information to the central government. The central government compared this information with the SISBEN information and determined whether a family was eligible for assistance or not. Thus, mayors had few opportunities to directly manipulate the magnitude of the program in their municipalities.
Notably, although the individual eligibility criteria and the criteria to determine the number of families covered by the program in each municipality was clearly established, there was not any explicit rule (formula) to determine the program’s coverage in each municipality. This could open the door for politicians to use this allocation for political purposes. As anecdotal evidence, local media and NGOs reported several complaints regarding potential FA program manipulation by the ruling national government during the 2010 Colombian presidential election. In its electoral observation mission report, Global Exchange claims that, of the 22 analyzed municipalities in which there were official complaints about program manipulation, the first round of the presidential election was won by one of the governing coalition candidates and the second round was won by the governing coalition candidate (with 74.6% of vote share on average). Moreover, the number of beneficiary families in the program in these municipalities grew from 38,214 in 2006 to 131,033 in 2010.
Figure 1 shows the evolution of the number of FA beneficiary families, and the percentage of municipalities with FA coverage in the country. The two waves of the FA program described above are clearly identified in this figure. At the beginning of the program, less than 30% of municipalities were covered, with only 84,000 families receiving benefits from the program. In 2009, almost 100% of the municipalities were covered by the program, with approximately 2,300,000 families receiving benefits. Consistently, central government spending on the program has exhibited a clear upward trend. In 2005, the FA program spending represented less than 0.3% of total government spending (nearly 200 million U.S. dollars). In 2009, FA program spending represented 3.5% of total central government spending (nearly 1.500.000.000 U.S. dollars). The FA beneficiary rate has exhibited a general increase over time, although it has also exhibited a high variability within and between municipalities.
4 Theoretical Framework and Identification Issues
To identify the political motives of the incumbent governing coalition underlying the allocation of the FA program across municipalities, we base our analysis on two of the main theories in political economics literature regarding the distribution of public projects across groups. Respective models can be divided in two: “loyal voter model” (Cox and McCubbins 1986) and “swing voter model” (Dixit and Londregan 1995, 1996; Lindbeck and Weibull 1987). Both models assume two parties competing to win an election by promising to distribute targetable goods to various groups. The main difference between these two models reflects the incumbent’s incentives to target different segments of voters.
On the one hand, the “loyal voter model” argues that the parties’ strategy is to allocate more public resources to “loyal” groups—that is, to those groups where the incumbent is electorally secure and expects positive electoral outcomes, but where an increase in the allocation of public resources is still necessary to mobilize core voters so as to positively affect the electoral results. On the other hand, the “swing voter model” argues that parties’ strategy is to allocate more public resources to swing groups—that is, to those groups where the incumbent is electorally vulnerable and expects negative electoral outcomes, but where an increase in the allocation of resources may favorably affect the electoral results. Clearly, these two models are not mutually exclusive. In other words, an incumbent may decide to target both swing groups and loyal groups at the same time.
We assume in our analysis that each municipality in the country is a group. Historically, many Colombian municipalities have exhibited strong political attachments with one of the two traditional parties (Conservador or Liberal). Although an important number of parties have emerged during the last two decades, as noted, most of them (and consequently, most of their voters) have preserved these political attachments. Additionally, some municipalities have historically been more willing to trade off their ideological attachments in exchange for a larger allocation of (either public or private) goods. Finally, as discussed above, the criteria to determine the number of FA beneficiary families in each municipality leave some room to manipulate the resources allocation across regions. Thus, to identify each municipality with a group makes sense in the Colombian context.
We hypothesize the following econometric model:
where i indexes municipalities, and t indexes time (with t = {2005, 2009}).5b is the FA beneficiary rate, l measures the concentration of loyal voters, and s mea sures the concentration of swing voters. Vector C includes variables that capture the preestablished criteria to determine the number of FA beneficiary families in each municipality. Equation 1 also includes the unemployment rate (u) as a control to investigate whether the incumbent governing coalition took into account this economic consideration when allocating the program. Finally, γbi is a municipal fixed effect, and μit is an idiosyncratic error term assumed to be uncorrelated with each other across municipalities. Parameters θ1 and θ2 in Equation 1 should be zero if the FA program was not politically manipulated by the governing coalition. If so, only parameters in φ must be statistically significant.
Estimates of equation 1 may suffer from reverse causality biases. For instance, increasing the spending in a municipality during a large period of time might increase the concentration of core voters in this municipality. In order to mitigate this problem, we consider historical measures for both the concentration of swing voters and the concentration of loyal voters, and compute these only considering electoral information before the expansion of the program. We formally define these measures in Section 5.
To identify the existence of political rewards resulting from the allocation of the FA program we hypothesize the following econometric model:
where i indexes municipalities, t indexes time (here with t = {2006, 2010}), v is the incumbent governing coalition vote share; X is a vector of other control variables (more on which below); γνi is a municipal fixed effect; and ϵit is an idiosyncratic error term assumed to be uncorrelated with each other across municipalities. We expect β1 to be positive and, depending on the voting behavior of loyal and swing voters, β2 and β3 could be positive, negative, or zero.
Introducing our political variables into equation 2 allows us to mitigate some potential bias in the estimation of β1. Since politicians could use the allocation of the programs for political purposes, if these political motives are not taken into account when estimating the effect of the allocation of subsidies on voting behavior, then the estimated effect is likely to be biased. Furthermore, since the criteria to determine the number of FA beneficiary families in each municipality (C ) might also affect the incumbent vote share, we include this set of variables as controls in equation 2. Finally, besides municipal fixed effects, we include a large set of control variables in vector X: economic, demographic, and violence controls, and the most important public expenditures in the country as executed by the central government across municipalities. We formally define all these measures in Section 5. Even so, estimates in equation 2 are likely to suffer from omitted variable bias. Unfortunately, there are not suitable instrumental variables available to address this problem. To address this problem, we conduct some placebo tests to provide additional confidence in our results.
5 Data and Measures
The unit of observation used in the analysis is the municipality. We concentrate on 2006 and 2010 presidential elections in Colombia. Our sample includes 1,028 municipalities (dictated by data availability); these represent 93% of the total municipalities in Colombia, and concentrate 98% of the total population in the country. The 7% of municipalities left out correspond to the states with the smallest populations in the country.
The number of FA beneficiary families in each municipality is taken from the Colombian government’s Social Action Office’s statistics. Using this information along with intercensus estimates of the population (computed by Colombia’s Statistics Office, DANE), we compute the number of FA beneficiary families (x 1,000 population). We refer to this as the FA beneficiary rate. Since presidential elections were carried out during the first semester of the corresponding year, this variable corresponds to the FA beneficiary rate one year before the election year (i.e., 2005 and 2009). We concentrate on this rate rather than on the ratio between the number of FA beneficiary families and the number of SISBEN 1 families because this measure takes into account the total population in a municipality and not only the number of poor, the first being the one that matters from a political point of view.
Information on the electoral results in each municipality is obtained from the Electoral National Office (Registraduria Nacional). As noted above, we consider two measures for the incumbent’s vote share, both of which differ only for the 2010 vote share that we take into account. The incumbent’s vote share (A) matches Uribe’s 2006 vote share with the first round sum of the governing coalition parties’ vote share for 2010. The incumbent’s vote share (B) matches Uribe’s 2006 vote share with Santos’s second-round vote share for 2010.
We use three measures of the concentration of loyal voters in a municipality. The first one is the average of the governing coalition’s share of votes in all races for president, mayor, and governor held during the preceding 8/9 years relative to the years under consideration. In particular, to compute the concentration of loyal voters in t = 2005/2006, we consider two races for president (held in 1998 and 2002), two races for mayor, and two races for governor (concurrently held in 2000 and 2003). Correspondingly, to compute this concentration of voters in t = 2009/2010, we consider two races for president (held in 2002 and 2006), two races for mayor, and two races for governor (concurrently held in 2000 and 2003). The other two measures are computed in a similar way but excluding or including some races during the same period. One of these measures takes into account only races for president and the other one takes into account races for president and the lower house (the Chamber of Representatives, concurrently held with presidential elections). Notably, considering these three measures allows us to explore the political interest that the incumbent governing coalition could have in different jurisdictional elections. Similar measures have been previously used by Stromberg (2004), Larcinese et al. (2006), and Ansolabehere and Snyder (2006).
As noted previously, Uribe’s party (Primero Colombia) was created for the sole purpose of supporting his presidential campaign. Moreover, with the exception of the Conservador party, the rest of parties belonging to the incumbent’s governing coalition were also created during the 2000s (after the 2002 election). Consequently, we do not have voting information for these parties prior to 2002. Therefore, for the years prior to 2002, we use only the share of votes obtained by the Conservador party in order to compute the concentration of loyal voters. For the elections held between 2002 and 2010, we consider not only the Conservador party’s share of votes, but also the share of votes of the parties belonging to the governing coalition that supported Uribe in these races (Partido de la U, Cambio Radical).
To measure the concentration of swing voters in a municipality, we use the standard deviation of the governing coalition’s share of votes over the preceding 8/9 years in all races for president, mayor, and governor. Consequently, we use the same electoral information used to compute the respective measure of loyalty. In a similar way, we also compute these variables by taking into account only races for president and by taking into account races for president and lower house. Thus, the concentration of swing voters in a municipality increases as the governing coalition’s vote volatility increases. This measure was suggested by Wright (1974), and has been used by several studies in this field. This swing measure is consistent with the Colombian electoral rules for all races, which is a majority one. Measures of pivotality, such as how near the vote share of a party is to a threshold, do not make sense under this electoral rule. The electoral information used in building our loyal and swing measures is obtained from the Electoral National Office (Registraduria Nacional).
As noted above, to mitigate potential reverse causality biases in equation 1, we calculate the concentration of loyal and swing voters by including electoral information only until 2006—that is, one year before the second (expansion) wave of the program. As explained in previous sections, although there was a small increase in the program coverage between 2003 and 2005, President Uribe did not pay much attention to this program during his first term in office, and only incorporated this as part of his platform in 2006.
We consider five variables to capture the municipal criteria to determine the number of FA families in each municipality. The first two variables are used to control for each municipality’s institutional capacity: the number of public teachers (x 1,000 school-age children), and the number of health care staff (x 1,000 population). Information about public teachers is taken from the Colombian Ministry of Education, and information regarding health staff is taken from the Colombian Ministry of Social Protection. The third variable is the number of SISBEN 1 families (x 1,000 population) in each municipality. As noted, this was considered to determine the priority in the allocation of subsidies across municipalities. The number of SISBEN 1 families is taken from the Colombian government’s Social Action Office’s statistics.6 The last two variables are used to control for the other two FA program’s constraints considered during the first wave: a dummy variable that takes value 1 if there is at least one bank office in the municipality, and population. Although, as we explained above, these variables were not included in the criteria to determine the allocation of subsidies across municipalities during the expansion wave, we decided to leave these variables in the econometric specification. Information regarding banks is taken from the Colombian Financial Superintendence.
To estimate equation 2, we include a large set of control variables in vector X. First, we include a political variable, which we call community council. During his two terms in office, almost every weekend, President Uribe organized a community council in a different municipality in the country. In these councils, he and some of his ministers discussed local necessities and problems with regional leaders and politicians. Thus, community council is a dummy variable that takes a value of 1 if during the 4 years leading up to the election there was in the municipality at least one community council; otherwise, it takes a value of 0. Although it is not clear how this variable might affect the incumbent’s vote share, it is plausible to expect that community councils positively affect this share. This variable is constructed using information from the Colombian Presidency webpage.
We also include economic, demographic, and violence controls. The economic variables included are per capita GDP, and the unemployment rate. The unemployment rate is taken from Colombia’s Statistics Office (DANE)7, and the per capita GDP is taken from the CEDE data set (the Center for Development Economics Studies, Universidad de los Andes). The demographic variables included are population and the percentage of female voters. This last variable is included to take into account differences in political preferences between females and males. These variables are taken from Colombia’s Statistics Office (DANE).
Since security has been a key political concern during the last decades in Colombia, in particular the internal civil conflict with guerillas, we also include three control variables into Equation 2 that take these issues into account. The first two variables represent the four year- average of the number of guerilla attacks (x 1,000 population) and the four year-average of the number of paramilitary attacks (x 1,000 population). We separate the attacks of these two groups because guerrillas have usually been associated with an extreme-left political ideology, whereas paramilitary groups have generally been associated with an extreme-right political ideology. We also control for the four-year average of the number of homicides (x 1,000 population). These variables are taken from the CEDE data set. Our expectation is that an increase in any of these rates will negatively affect the incumbent’s vote share.
Finally, we add to this econometric specification the most important public expenditures in the country as executed by the central government across municipalities. Since the allocation of these resources might be correlated with allocation of the FA program, including these variables allows us to avoid potential omitted variable biases. We include the per capita amount of real transfers from the central government to each municipality. Although the distribution of these transfers across municipalities is done on the basis of an established rule—one that takes into account such dimensions as population and poverty—these transfers represent the most important source of revenues for the municipalities. We also include the per capita spending allocated by Colombia’s Family Welfare Institute (ICBF, its acronym in Spanish) to each municipality. The ICBF is a central government agency that aims to protect families, especially children, in the country. This program is one of the largest social programs in the country run by the central government. Finally, we also include the number of displaced families with FA coverage (x 1,000 population). As explained above, in addition to resources allocated through the FA program to the poorest families, the Social Action Office allocates resources to displaced people in the country. Transfers are taken from the CEDE data set, ICBF spending was supplied by the ICBF, and the FA displaced population beneficiary rate is constructed using information from the Social Action Office.
Descriptive statistics of these variables are reported in the Appendix (Table 1A). Note that in 2006, there are some municipalities where the measures of swing and loyal both have a value of 0. Actually, this only happens in one municipality, where the incumbent governing coalition both failed to field a candidate for local elections and failed to receive a single vote in the presidential race.
6 Results: Incumbent Allocation Strategy
In this section, we present our estimates of the determinants of the allocation of the FA program across municipalities (equation 1). The dependent variable under consideration here is the FA beneficiary rate. The main variables of interest are the historical measures of the respective concentrations of swing voters and loyal voters introduced above. If there have not been political motives behind the allocation of the program, we would expect that only those variables accounting for the allocation criteria of subsidies across municipalities would have explained the final geographical allocation of the program over time.
Table 1 reports our results. All the regressions include time effects in order to take into account the general upward trend of the FA beneficiary rate. We report estimates for the three sets of loyal and swing measures decribed earlier. Column 1 reports the estimates combining the measures for the concentration of loyal voters and swing voters in races for president, mayor, and governor; column 2 those combining the same measures in races for president; and column 3, those combining the same measures in races for president and lower house.
Allocation of the FA Program and Incumbent’s Allocation Strategy Fixed-Effects Estimates Dependent variable: FA beneficiary rate
Robust standard errors in parentheses. ***1%,** 5%, and *10% significance. All regressions include time-fixed effects.
Allocation of the FA Program and Incumbent’s Allocation Strategy Fixed-Effects Estimates Dependent variable: FA beneficiary rate
Robust standard errors in parentheses. ***1%,** 5%, and *10% significance. All regressions include time-fixed effects.
The results reported in columns 1 and 2 of Table 1 strongly support the swing voter hypothesis. We find a positive and statistically significant effect of the concentration of swing voters on the allocation of the FA program. This effect is greater when considering the swing and loyal measures in column 1. An increment of one standard deviation in the concentration of a municipality’s swing voters implies an increment of 6.5 points in the FA beneficiary rate (13% of the mean). The respective effect when considering the result in column 1 is 3.1 (6% of the mean). This effect becomes statically irrelevant when considering the respective swing and loyal measures in column 3. This suggests the likely interest of the governing coalition in affecting the electoral results in the 2007 races for mayor and governor, and in the 2010 race for president.
The results reported in Table 1 provide just the opposite evidence for the concentration of loyal voters. We find a negative and statistically significant effect of the concentration of loyal voters on the allocation of the FA program. Once again, this effect is greater when considering the respective swing and loyal measures in column 1, and disappears when considering the respective measures in column 3.
Regarding the effect of the institutional variables, we find that public teachers and health care staff positively affected the allocation of resources through the FA program—as dictated by the program’s criteria. Furthermore, the number of FA families also exhibited a positive relationship with the FA beneficiary rate. The existence of bank offices had no effect on this allocation, although the total population did. Our estimates also show that the FA program was more intensively expanded in those municipalities that exhibited a greater increase in the unemployment rate. This can be also understood as a manipulation of the program allocation.
In sum, our results show that although the governing coalition took into account the program’s expansion criteria for the distribution of resources across municipalities between 2005 and 2009, this targeted more extensively those regions that were more likely to change their votes in favor of the incumbent. This result is consistent with the political record of former president Uribe. Most of his political capital has been concentrated in the states belonging to the so-called “coffee region” of the country (in particular Antioquia—where Uribe was born—Caldas, Risaralda, and Quindío), and in some states located in the middle southeast of the country, all of them with strong political attachments with the Partido Conservador. Today, these regions are still highly loyal to Uribe. Thus, from a strategic point of view, it is quite likely that the governing coalition used the FA program to target intensively swing regions in order to extend Uribe’s popularity throughout the country—even more if Uribe was looking for a second reelection, and that winning more political support in the 2007 mayor and governor elections could be useful for this purpose. Interestingly, our results are totally different from those reported in Fried (2012), who finds that there is a weak relationship between the allocation of the CCT program in Brazil and political variables.
7 Results: Political Rewards
In this section, we study whether or not the incumbent’s governing coalition was rewarded politically for the subsidies allocated through the FA program. Table 2 reports our estimates. In columns 1 and 2, the endogenous variable is the incumbent’s vote share (A), while in columns 3 and 4, the endogenous variable is the incumbent’s vote share (B). Columns 1 and 3 report the effect of the FA program on the respective incumbent’s vote share when the swing and the loyal measures are omitted from the econometric specification, and columns 2 and 4 report this effect when both variables are included. Our purpose in doing this is to show how the estimation for political rewards changes as the political motives of the incumbent governing coalition underlying the allocation of the program are omitted. We use the loyal and swing measures that take into account races for president, mayor, and governor in all regressions reported in Table 2. The results are qualitatively the same when we estimate Equation 2 using the other two sets of measures.8
Regardless of the incumbent vote share under consideration, we find a positive and statistically significant effect of the FA beneficiary rate on this. Moreover, for each of the incumbent vote share measures, the magnitude of the estimated effect is quite similar regardless of whether the loyal and swing measures are included or not in the econometric specification. These results partially show how the estimation of political rewards is quite robust to the inclusion of relevant variables that are important in explaining the incumbent vote share.
The estimated effect of the FA beneficiary rate on the incumbent’s vote share (B) is larger than the estimated effect of this rate on the incumbent’s vote share (A). An increment of one standard deviation in the FA beneficiary rate in a municipality positively affected vote share (A) in 1.96 percentage points (3% of the mean), and vote share (B) in 2.94 percentage points (4.2% of the mean). The difference in the magnitude of these effects may be due to the fact that the competition between the incumbent and the challenger was more intense during the second round.
Consistent with the allocation strategy of the FA program discussed above, our results show that the incumbent governing coalition became more successful in capturing votes in historically swing municipalities, and less successful in historically loyal municipalities. Although these effects can be explained by other factors, an important part of these can be understood as an indirect effect of the allocation of the program on the incumbent’s vote share.
Taken together, these results suggest that voters rewarded the incumbent governing coalition not only for redirecting the subsidies of the program strategically but also for the allocation of subsidies in itself. This is based on the fact that the effect of the FA program on the incumbent vote share is positive and statistically significant even after controlling for political variables that affected its allocation across municipalities.
Effect of FA Program on Incumbent Vote Share Fixed-Effects Estimates
Robust standard errors in parentheses. ***1%,** 5%, and *10% significance. All regressions include time-fixed effects.
Effect of FA Program on Incumbent Vote Share Fixed-Effects Estimates
Robust standard errors in parentheses. ***1%,** 5%, and *10% significance. All regressions include time-fixed effects.
Placebo Tests: Effect of FA Program on Incumbent Vote Share Fixed-Effects Estimates
Robust standard errors in parentheses. ***1%,** 5%, and *10% significance. All regressions include time-fixed effects. Regressions include the same controls used in columns 2 and 4 of Table 2. These parameters are not reported.
Placebo Tests: Effect of FA Program on Incumbent Vote Share Fixed-Effects Estimates
Robust standard errors in parentheses. ***1%,** 5%, and *10% significance. All regressions include time-fixed effects. Regressions include the same controls used in columns 2 and 4 of Table 2. These parameters are not reported.
There are some interesting results regarding the effect of other control variables included in Table 2. First, the incumbent vote share was negatively affected in those municipalities where the number of SISBEN 1 families increased more. Since the number of SISBEN 1 families should be positively correlated with the poverty level in each municipality, this suggests that the incumbent governing coalition became less successful in capturing votes in those municipalities where the economic conditions worsened. Nevertheless, this might also reflect some discontent in those regions where, despite the existence of a large number of eligible families, the expansion of the program was not large enough to meet their demands.
Finally, those municipalities that suffered an increase in guerilla attacks more intensively supported the incumbents’ coalition. This result reflects the success of the main political platform of the governing coalition—that is, the fight against guerrillas. Since guerrillas were forced to redirect their attacks due to the government’s military strategy, those municipalities that suffered an increase in guerrilla attacks were more likely to support the incumbent’s coalition based on its success in other municipalities.
Table 3 reports two additional tests that aim to show the robustness of the estimations for political rewards reported in Table 2. We only report the parameters of interest in this table. Column 1 (Test 1) reports the results of running the same regression reported in columns 2 and 4 of Table 2, but using as the dependent variable the incumbent vote share in 2002 and 2006 rather than the respective share in 2006 and 2010. Remember that there was only one round in the 2002 and 2006 presidential elections. The variable of interest in this case is still the FA beneficiary rate in 2006 and 2010. Hence, Test 1 allows us to infer whether estimates in Table 2 are omitting relevant variables regarding municipal preferences for the incumbent governing coalition or not. If this was the case, we should find that the FA beneficiary rate between 2006 and 2010 affected the incumbent vote share one period before. We do not find evidence that this was the case.
Columns 2 and 3 (Test 2) report the results of running the same regressions reported in columns 2 and 4 of Table 2, using the same dependent variables (incumbent vote share A and B respectively), but including as exogenous variable the FA beneficiary rate in 2002 and 2005 rather than the respective rate in 2005 and 2009. Hence, Test 2 allows us to infer whether the results reported in Table 2 are driven by the implementation of the program during its first wave rather that by the expansion of the FA program during its second wave. We also find that this was not the case.
8 Conclusions
We have reported evidence concerning both the political manipulation of the CCT program in Colombia, and the political rewards that the incumbent governing coalition obtained by providing this program. In particular, we find that although the program’s institutional criteria were taken into account for the distribution of resources across municipalities, the governing coalition targeted more intensively swing regions and less intensively loyal regions. We understand this result as evidence of political manipulation of the program. From a strategic point of view, it is quite likely that the governing coalition had decided to target intensively swing municipalities with the FA program while supplying public goods, like the security policy against guerrillas, in loyal municipalities—where this policy was highly valued.
We have also presented evidence supporting the hypothesis that incumbent politicians derive political reward through the allocation of subsidies via large CCT programs. Notably, our results suggest that voters rewarded the incumbent governing coalition not only for redirecting the subsidies of the program strategically but also for the allocation of subsidies in itself. This result is consistent with the notion that voters highly value government’s cash transfers in countries with high levels of poverty and inequality.
Our joint results suggest that CCT programs in developing countries may be used by incumbents in order to increase their political support. While it is almost impossible (and also undesirable) to avoid the existence of political rewards from the implementation of policies that voters want (like CCT programs in developing countries), it is desirable, for redistribution reasons, to avoid the political manipulation of these programs. In order to do this, the allocation of these resources should respond better to technical criteria from a geographical perspective. Potential mechanisms for improving the geographical allocation of the program’s resources while minimizing political manipulation should be implemented.
Appendix
Summary Statistics for Municipality-Level Data
The information of variables indicated with (₤) correspond to 2005 and 2009 respectively. Incumbent vote share 2010 (A) corresponds to the first-round governing coalition’s vote share. Incumbent’s vote share 2010 (B) corresponds to the 2010 second-round Santos’s vote share. In 2006, there are some municipalities where the measures of swing and loyal both have a value of 0. Actually, this only happens in one municipality, where the incumbent governing coalition both failed to field a candidate for local elections and failed to receive a single vote in the presidential race.
Summary Statistics for Municipality-Level Data
The information of variables indicated with (₤) correspond to 2005 and 2009 respectively. Incumbent vote share 2010 (A) corresponds to the first-round governing coalition’s vote share. Incumbent’s vote share 2010 (B) corresponds to the 2010 second-round Santos’s vote share. In 2006, there are some municipalities where the measures of swing and loyal both have a value of 0. Actually, this only happens in one municipality, where the incumbent governing coalition both failed to field a candidate for local elections and failed to receive a single vote in the presidential race.
Notes
We report 2009 data because this is the last year covered by our sample.
Source: Colombian Department of National Planning, hereafter DNP for its acronym in Spanish.
Although Camacho and Conover (2011) documented political manipulation of this index before 2003, no evidence of manipulation is known during our period of analysis.
See Acción Social and DNP (2010) for a more extended description of the program.
Since presidential elections were carried out during the first semester of the corresponding year (2006 and 2010), we analyze the FA beneficiary rate one year before the election year.
Because of data availability limitations, for 2005 we use the number of SISBEN 1 families in 2006.
Unemployment rate information is only available at the state level.
The estimations for political rewards are qualitatively the same. The estimated effects of the concentration of loyal voters and swing voters on the incumbent vote share are qualitatively the same when using respective measures that take into account races for president. We find no effect of these variables on the incumbent vote share when using the measures that take into account races for president and lower house. These results are available upon request.
