ABSTRACT
While the literature has documented negative effects of conflict on educational outcomes, there is little evidence on the effect of conflict termination. We show how the permanent ceasefire declared by FARC’s insurgency during peace negotiations with the Colombian government caused a differential improvement on several educational outcomes in the areas affected by FARC violence prior to the ceasefire. These effects are not explained by peace building and post-war recovery investments, and they are only partially driven by wartime child soldiering. Instead, we find support for other mechanisms, such as the post-ceasefire plummeting of victimization and new economic opportunities in treated areas.
I. Introduction
Civil war is an enormous obstacle to development as it entails large economic and social costs (see, for example, Goldin and Lewis 1975; Collier 1999; Abadie and Gardeazabal 2003). Of these, perhaps the most important is the loss in human capital because of its effect on long-term labor market and health outcomes (Mincer 1974; Almond, Currie, and Duque 2018; Barker 1998; Cunha and Heckman 2007). A lower productivity, in addition, decreases the opportunity cost to engage in illegal activities (Becker 1968), thus triggering a long-run vicious cycle of violence and lack of opportunities, even after conflict has ended (Justino, Leone, and Salardi 2014; Unesco 2011; Leon 2012; Duque 2017).1
But can the end of a conflict counteract, at least partially, the human capital loss generated by violence? The empirical evidence on this is rather scarce. We study the effect of the recent efforts to end the five-decade-long conflict in Colombia on a number of educational outcomes and conclude that the end of the conflict generated sizeable short-term gains in terms of human capital accumulation. Specifically, we find that the permanent ceasefire declared by the Revolutionary Armed Forces of Colombia (FARC from the Spanish acronym) in the context of a peace process with the Colombian government caused significant reductions in dropout rates, as well as grade failure rates and increases in high school graduation and test scores in municipalities formerly affected by FARC violence relative to other areas.2
Conceptually, these findings are far from obvious. On the one hand, when violence stops the conditions for children to return to school, remain at school, graduate from school, and perform better in school tests may be more favorable due to the generalized perception of safety, the potentially better economic prospects, and the reintegration into the schooling system of former child soldiers. On the other hand, war is destructive, and targeted infrastructure may include school facilities and the road network used by children to attend schools. Also, the returns from schooling may be low after conflict if violence disrupts markets and exchange. Finally, winning parties may prevent the defeated side to access formal education by creating racial, ethnic, or religious requirements for enrollment (see Bush and Saltarelli 2000; Shemyakina 2011). Surprisingly, however, how the end of conflict affects educational outcomes in the short run has been seldom studied. The available evidence is at best indirect, as studies that focus on the impact of short conflicts implicitly address the effect of post-conflict recovery, especially when comparing exposed and nonexposed cohorts (for example, Dabalen and Paul 2014).
This contrasts sharply with the extensive literature that provides abundant and compelling evidence that conflict hurts human capital accumulation. Through the occurrence of killings, injuries, displacement, trauma, and disease, civil war causes large reductions in both the stock of human capital and its growth rate. In a recent review of the mounting subnational evidence, Justino (2016) separates the mechanisms of the effect of conflict on human capital into supply and demand channels. Supply channels include the destruction of infrastructure, social capital and markets, the depletion of financial resources, and teacher victimization and absenteeism.3 Demand channels include child labor (used to cope with war-driven impoverishment or to replace household labor due to death, injury or recruitment), poor health conditions resulting from conflict exposure (including malnourishment, stress during pregnancy, and psychological trauma), and child soldiering.4
Our estimates of the effect of the end of the conflict with FARC are large and robust. Specifically, we find that after the ceasefire (2015–2018), municipalities previously exposed to FARC violence experienced a differential 19 percent reduction in dropout rates, 13 percent reduction in grade failure rates, and 1.5 percent and 1.8 percent increases in math and reading test scores, respectively. Importantly, we also find a differential 19 percent increase in high school graduation. According to the available estimates for Colombia, the latter implies differential wage returns of 23 percent and 26 percent in rural and urban areas, respectively (Vargas-Urrutia 2013). Thus, the short-term productivity gain of the end of the conflict with FARC is potentially quite large. These results are robust to using different measures of exposure to FARC violence and to using as control areas only municipalities affected by violence perpetrated by other armed groups or municipalities matched in terms of several pre-ceasefire characteristics. They are also robust to the inclusion of department-by-year fixed effects and to controlling for differential changes in educational outcomes after the ceasefire parametrized by various pre-ceasefire municipality characteristics.5
We also explore the potential mechanisms that relate the end of violence with a generalized short-term improvement in educational outcomes. While we find that the recruitment of children decreased in formerly FARC-affected areas relative to the rest of the country, we find no heterogeneous effects in terms of characteristics that correlate with the typical profile of most illegally abducted children—gender, age, and the urban versus rural location of schools. Moreover, a back-of-the-envelope calculation suggests that child recruitment can only explain a small fraction of the estimated reduction in dropout rates (up to 8.5 percent). Instead, our evidence is consistent with other potential mechanisms, namely the large post-ceasefire reduction in victimization and the expectation of better economic prospects. On the first, we find that the positive effects of the ceasefire on educational outcomes are larger in places that experienced more pre-ceasefire violence, victimization from land mines, and internal forced displacement.6 On the second, we document a differential increase in the number of new firms during the first two years after the start of the ceasefire and show that the improvement of educational outcomes is larger in the areas that experienced this surge. In contrast, in areas featuring economic opportunities that are substitutes of schooling—because of their intensity in unskilled and child labor—the effect of the ceasefire on educational outcomes is attenuated. This is the case of places that are highly suitable to grow coca bushes (used to produce cocaine). This is important because it suggests that, by increasing the opportunity cost of schooling, some economic activities can reduce the human capital dividend of conflict termination.
This work contributes to several strands of the literature. First, as mentioned, educational outcomes would not necessarily respond symmetrically to conflict and to the end of it. The empirical literature has focused on the effect of conflict violence, with particular attention on long-term impacts. However, the evidence on how conflict termination affects school attainment, school completion, test scores, and human capital accumulation in general (in either the short or the long-run) is scarce. As violence-affected countries move forward in the process of transitioning to peace, it is crucial to understand how individuals in affected areas respond to the absence of violence. Second, while previous studies have focused on the post-conflict welfare of either ex-combatants (Blattman and Annan 2010) or veterans (Angrist 1989), besides this paper there is little evidence on the welfare of civilians, especially children. Third, we contribute to the study of the consequences of the end of the Colombian conflict with FARC. Other papers highlight important unintended negative consequences in terms of the security of local leaders (Prem et al. 2020) and deforestation (Prem, Saavedra, and Vargas 2020). In light of this evidence, the finding that educational outcomes largely improve following the ceasefire provides a silver lining.
The rest of the paper is organized as follows. Section II provides some background information, and Section III summarizes the data sources. Section IV describes the identification strategy to estimate the effect of the ceasefire on educational outcomes, and Section V reports the main findings and robustness. Section VI investigates the potential mechanisms behind our main results, and Section VII concludes.
II. Context
A. Colombia’s Education System
The education system in Colombia comprises one year of preschool, five years of primary education, four years of lower secondary education, and two years of upper secondary education. In 2014, 87 percent of the schools in Colombia were public, and out of those, 78 percent were located in rural areas (OECD 2016). All children between five and 15 years old are legally required to attend preschool plus nine years of compulsory basic schooling. However, it is estimated that 20 percent of the students do not continue studying beyond primary school (OECD 2016), and only 65 percent (77 percent) of boys (girls) complete lower secondary education (Radinger et al. 2018).
One of the main factors associated with early school dropout and the failure to graduate is violence exposure (García, Monsalve, and Torres 2010). Unilateral attacks as well as bilateral clashes between armed groups threaten families and communities, and, according to OECD (2016), school-age children are more likely than other age groups to be affected by violent death, recruitment, and displacement. Indeed, Rodríguez and Sánchez (2010, 2012) show that armed conflict reduces educational attainment and decreases the academic achievement of students that attend schools in conflict-affected areas. In addition, Fergusson, Ibañez, and Riaño (2019) find that individuals exposed to intense violence during the “La Violencia” civil war that took place in the 1940s and 1950s achieved up to 0.3 fewer years of education and were more likely to work in less productive sectors as adults.
B. Colombia’s Civil War and the Peace Process
Colombia’s civil conflict started with the foundation of left-wing guerrillas FARC and the National Liberation Army (ELN from the Spanish acronym) in the mid-1960s. Guerrillas claim to represent the rural poor and have fought for more than 50 years with the stated aim of overthrowing the government. In order to finance the protracted war, both groups have been profiting from several forms of illegal activities localized within the Colombian territory (Richani 1997). This implies that subnational territorial dominance is an important intermediate objective of the armed groups, and the infliction of violence on both military and civilian targets is a means of achieving it (Kalyvas 2006).
The conflict was a Cold War proxy until the end of the 1980s, but escalated during the 1990s fueled by the guerrillas’ involvement in illegal drug trafficking and the consolidation of antiguerrilla right-wing paramilitary groups. In the mid-1990s, the paramilitaries effectively became a third force in the conflict when splintered paramilitary armies colluded under the umbrella organization of the United Self-Defense Groups of Colombia (AUC by its Spanish acronym). The five-decade-long, three-sided Colombian conflict resulted in almost 8.8 million people formally registered with the state as victims of the conflict.7
In October 2012, the Colombian government and FARC started peace negotiations in Cuba. While the four-year-long process was characterized by a constant ebb and flow, one of the most significant milestones was the establishment of a permanent ceasefire by FARC on December 20, 2014. In fact, as a result of the ceasefire, FARC withdrew their troops to more remote areas where military contact with government security forces and other armed groups was unlikely to take place. This explains why FARC’s offensive activities dropped by 98 percent during this period (CERAC 2016). To further illustrate this point, Table 1 reports the change in the average number of violent events in municipalities exposed to FARC violence before (2009–2014) and after the ceasefire (2015–2018). The table shows a systematic reduction in violence in all categories in municipalities formerly exposed to FARC after the ceasefire. For example, the number of war-related actions dropped by 79 percent, and the victims from antipersonnel mines plummeted by 69 percent.
Changes in Conflict
We show that FARC’s inability to exert violence by their own initiative or to respond violently to actions perpetrated either by the military or other illegal armed groups during the ceasefire (which was largely met until replaced by the bilateral definitive ceasefire and the subsequent disarmament of FARC in 2016) generated a sizable reduction in the incidence of violence in municipalities previously exposed to FARC. In turn, this resulted in a sizable improvement in a wide range of educational outcomes and thus in a differential increase in children’s human capital accumulation.
III. Data
We built a municipality–year panel to study the effect of the permanent ceasefire on educational outcomes (Prem, Vargas, and Namen 2021). We focus on 2009–2018, which includes the last year of the presidential term of Álvaro Uribe, the whole presidential term of Juan Manuel Santos, and the first year of the presidential term of Iván Duque.8 Our sample consists of 1,092 municipalities with a population of less than 200,000. We drop from our sample large cities and department capitals, which are less affected by conflict and largely urbanized.9 We now describe the main variables and their sources.
A. Education Data
We study the effect of the ceasefire on outcomes related to schooling and academic achievement. The former include dropout and grade failure rates for all school years, as well as high school graduation. To construct these we rely on the Colombian school census (officially called Form C-600), which is collected yearly by the Statistics Bureau (DANE from its Spanish acronym) and the Ministry of Education. These data are reported at the school–grade level and are disaggregated by gender. The school census also contains information on whether schools are located in rural or urban areas and on whether they are public or private. For both dropout and failure rates, we compute a municipality-level weighted average of the school-specific rates, using as weights the share of the school-level enrollment over the municipality’s entire school population. In turn, dropout rates at the school level are computed as the share of students that leave a school during the academic year and do not continue their studies, relative to the initial school enrollment.10 Likewise, school failure rates are computed as the share of students who do not approve the school year and hence fail to be promoted to the next grade, relative to the same denominator. Municipal-level high school graduates are computed as the logarithm of the number of students promoted the last year of high school, added across all the municipality schools.
The outcomes associated with academic achievement are the reading and math test scores obtained in standardized national exams (called “Saber”) that are implemented yearly in Grades 3, 5, and 9. This information comes from administrative data sets of the Colombian Institute for the Evaluation of Education (ICFES from its Spanish acronym) and is available for 2012–2017 in the form of school-level averages.11 We compute a municipality-level weighted average for each grade/subject using as weights the number of students that took the test in each school–year.
B. Conflict Data
To construct a measure of exposure to FARC violence prior to the start of the ceasefire, we use the conflict data set originally compiled by Restrepo, Spagat, and Vargas (2004) and updated through 2018 by Universidad del Rosario. This data set codes violent events recorded in the Noche y Niebla reports from the NGO Centro de Investigación y Educación Popular (CINEP) of the Society of Jesus in Colombia, which provides a detailed description of the violent event, its date of occurrence, the municipality where it took place, the identity of the perpetrator, and the count of the victims involved in the incident.12
Our main measure of exposure to FARC violence is an indicator that identifies municipalities as highly exposed to FARC violence. This dummy takes the value one for municipalities in the top three quartiles of the empirical distribution of the total number of FARC attacks (normalized by 10,000 inhabitants) that took place from 2011 to 2014.13 By dropping the bottom quartile, we do not consider the municipalities with few attacks as highly exposed. Our results are robust to using an alternative indicator of “high exposure” based on the top half of the empirical distribution of total FARC attacks, as well as the continuous measure. Figure 1 portrays the spatial distribution of the continuous measure.
Distribution of Exposure to FARC Violence
Notes: This figure presents the spatial distribution of attacks per capita by FARC previous to the ceasefire. Darker gray shades represent municipalities more affected by FARC violence prior to the ceasefire.
C. Other Data
We complement these data with a range of municipal characteristics, including tax income, expenditures, transfers from the central government, average crops’ yields, and the share of agricultural land. These variables come from an annual panel of Colombian municipalities constructed by the Center for the Study of Economic Development from Universidad de los Andes. We also include official statistics on infant mortality rates and data on firm creation from the Colombian Confederation of Chambers of Commerce.
D. General Pre-Ceasefire Patterns
Online Appendix Table A.1 reports, for the pre-ceasefire period (2009–2014), summary statistics for all the variables used in all the regression models that we estimate. The baseline yearly dropout and failure rates are 5 percent and 8.2 percent, respectively. Both are somewhat higher in public schools, in secondary school grades, and for boys. However, dropout rates are higher in rural areas, while failure rates are higher in urban areas. Most high school graduates come from urban and public schools, and the majority of them are girls.
Regarding the treatment variable, during 2011–2014, 7 percent of the country was highly exposed to FARC violence. Online Appendix Table A.2 shows that, during this period, there are level differences in all the educational outcomes, with nonexposed municipalities enjoying better outcomes. In spite of these differences, Figure 2 suggests that the pre-ceasefire trends in most of the outcomes are quite similar across exposed and nonexposed areas. We will test this similarity more formally in order to assess the empirical validity of our main identification assumption. To complement the absence of pre-trends in the outcome variables, Figure 3 shows that various other key municipal characteristics that are likely to affect educational outcomes also lack differential pre-ceasefire trends. These include municipal tax income and expenditure, transfers from the central government, agricultural activity, and firm entry.
Raw Data Dynamics for Educational Outcomes
Notes: This figure presents the evolution of municipality education outcomes for exposed and nonexposed municipalities to FARC violence.
Raw Data Dynamics for Other Municipality Characteristics
Notes: This figure presents the evolution of municipality characteristics for exposed and nonexposed municipalities to FARC violence.
Finally, to motivate the subsequent formal empirical analysis, Online Appendix Figures A.1–A.3 show the relationship between the change in each of the educational outcomes before and after the ceasefire and the indicator of high exposure to FARC violence, as well as with the continuous exposure measure. These figures suggest that municipalities with more exposure to FARC violence prior to the ceasefire have larger decreases in dropout rates and failure rates, as well as larger increases in high school graduation rates and test scores. In the rest of the paper, we study these suggestive patterns more formally.
IV. Empirical Strategy
A. Main Specification
Our identification strategy exploits the timing of the permanent ceasefire announced by FARC on December 20, 2014, as well as the spatial distribution of the exposure to FARC violence across municipalities prior to the ceasefire. More formally, using the subindex m to denote municipalities, d to denote departments, and t to denote year, we estimate the following difference-in-differences model:
1
where ymdt is any of our schooling or academic achievement outcomes, FARCm measures pre-ceasefire high exposure to FARC violence in municipality m, and Ceaset is a dummy that takes the value one after the start of the permanent ceasefire (that is 2015–2018). αm are municipality fixed effects, and λdt are department-by-year fixed effects. These control, respectively, for any observed or unobserved municipal-level time-invariant heterogeneity and for any time shocks that affect all the municipalities of the same department simultaneously. Xm are municipality characteristics measured before the ceasefire that we interact with the ceasefire time indicator to flexibly control for differential changes pre- and post-ceasefire parametrized by each one of the municipal attributes. The set of characteristics includes the logarithm of population, the share of rural population, a poverty index, and the distance to the department capital. Finally, εmdt is the error term, which we cluster at the municipality level.14
All regressions are weighted by the number of students enrolled in 2014 (the last pre-ceasefire year) in each municipality. In this way, we give the same weight to every student, and thus our coefficient of interest, β, captures the differential change before and after the ceasefire in educational outcome y in municipalities highly exposed to FARC violence relative to the rest of the country. Our results are, however, robust to using other weights or to not weighting the observations at all, thus giving the same importance to every municipality regardless of the student population.
B. Identifying Assumption
The main assumption behind our difference-in-differences model is that in the absence of the ceasefire, educational outcomes would have evolved similarly in municipalities highly exposed to FARC violence and in municipalities not exposed to FARC violence. The validity of this “parallel trends” assumption can be partially assessed by estimating the following equation:
2
where δj are year dummies, and T includes all years in our sample except 2014, which is the year before the start of the ceasefire. Therefore parameters βj can be interpreted as the difference in outcome y in municipalities exposed to FARC violence and nonexposed municipalities, in year j relative to the year at the end of which the ceasefire started.
C. Potential Mechanisms
We can use variation across student, school, or municipal-level characteristics to estimate heterogeneous effects that may illustrate some of the underlying mechanisms. In particular, any change in educational outcomes after the start of the ceasefire may be explained by the generalized decrease in the victimization of civilians in places previously affected by FARC violence, by pre-ceasefire child recruitment perpetrated by FARC, or by newly available economic opportunities (that can be either complements or substitutes of schooling). We thus divide a set of potential mechanisms into these categories and test whether the effect of the ceasefire on educational outcomes vary across these dimensions. To do so, we augment the main specification in Equation 1 by adding a third interaction term. Denoting the pre-ceasefire student/school/municipal potential mechanism as Zs/m, we estimate:
3
Our coefficient of interest, β1, captures the differential change in schooling and academic achievement outcomes in places highly exposed to FARC violence for students/schools/municipalities with characteristic Zs/m. The set of characteristics Zs/m includes differences by gender, primary/high school grades, rural/urban location, public/private ownership, exposure to other armed groups, exposure to land mines, experience of forced displacement, firms’ creation, and coca suitability. Note that the results coming from this test are suggestive about potential mechanisms but not necessarily causal, so they have to be interpreted with caution.
Using the above specifications we estimate the impact of the December 2014 permanent ceasefire on educational outcomes in areas previously exposed to FARC violence (Equation 1), the dynamic persistence of this effect (Equation 2), and heterogeneous effects (Equation 3). The next section reports the estimated results.
V. Results
This section presents our main findings related to both the schooling outcomes and the outcomes associated with academic achievement. We also provide evidence consistent with the main identifying assumption of our empirical strategy and discuss further robustness tests.
A. Schooling Outcomes
In Table 2 we report the coefficients resulting from estimating Equation 1 on three key schooling outcomes: dropout rates (Columns 1 and 2), grade failure rates (Columns 3 and 4), and high school graduation (Columns 5 and 6). All columns include municipality as well as department-by-year fixed effects. In addition, even columns include a large set of predetermined municipal controls interacted with the ceasefire dummy. The standard errors in parentheses are clustered at the municipality level. For robustness, in square brackets we report p-values that take into account the potential cross-sectional dependence of the error term (Conley 1999, 2016).15 Table 2 also reports the p-values of standard errors corrected for multiple hypotheses testing.16
Educational Outcomes, Exposure to FARC Violence, and Ceasefire
We find that, relative to the rest of the country, places that were highly exposed to FARC violence experienced, after the start of the ceasefire, lower dropout and failure rates and higher high school graduation. We compute the economic size of the effects based on the most demanding specification, which includes differential municipal trends parametrized by several pre-ceasefire municipality characteristics. This is a conservative choice, since adding these controls slightly reduces the size of the estimated coefficients for the three outcomes.17 After the start of the ceasefire, municipalities traditionally highly exposed to FARC violence experienced a relative decrease in dropout rates equivalent to 19 percent of the pre-ceasefire mean and 32 percent of the standard deviation. Likewise, failure rates decreased in a magnitude equivalent to 13 percent of the pre-ceasefire mean and 36 percent of the standard deviation. Finally, the number of high school graduates increased by 19 percent.
B. Academic Achievement Outcomes
Table 3 follows the same structure as Table 2 for the academic achievement outcomes. Panel A reports the results related to the performance in the math module of standardized school tests. Panel B focuses on the performance in the reading module. As explained in Section III, in Colombia, national standardized tests take place in Grades 3, 5, 9, and 11. Because of data comparability issues over time, we focus on the first three. In addition to the standard errors clustered at the municipality level and the p-values of standard errors that allow for cross-sectional dependence.
Learning Outcomes, Exposure to FARC Violence, and Ceasefire
Based on the most demanding specification, we find that after the start of the ceasefire, math tests scores differentially improved in places highly exposed to FARC violence in 1.6 percent of the pre-ceasefire mean and 15 percent of the standard deviation. This magnitude is the same for the test results in all the available school grades. Regarding reading test scores, we find an increase in 1.2–1.8 percent of the pre-ceasefire mean and in 11–18 percent of the standard deviation.
C. Identifying Assumption
In this subsection we assess the validity of our empirical strategy by showing evidence consistent with its main identifying assumption. First, Figure 4 plots the coefficients coming from estimating Equation 2 for our nine educational outcomes. Panels A–C report schooling outcomes, and Panels D–I report those associated with academic achievement. With the exception of math test scores in the ninth grade national standardized test, it can be seen that before the ceasefire, the estimated coefficients are not statistically significant and, in most cases, are close to zero. This points to the absence of differential pre-ceasefire trends in most educational outcomes when comparing municipalities that were highly exposed to FARC violence and places that were not. Thus, this provides support for using a difference-in-differences empirical strategy to estimate the effect of the ceasefire on human capital accumulation. In turn, except for grade failure rates and the log of high school graduates (Panels B and C, respectively), the estimated coefficients are significantly different from zero—and move in the expected direction—after the start of the ceasefire.
Dynamic Specification for Educational Outcomes
Notes: This figure presents the coefficients from our dynamic specification presented in Equation 2 for our main educational outcomes. We present the point estimates of the regression and the confidence of interval at 95 percent.
We also conduct a parametric test for the existence of differential trends during the pre-ceasefire period (2009–2014) in the spirit of Muralidharan and Prakash (2017). We do so by interacting a linear trend with our measure of exposure to FARC violence and test for the significance of the associated coefficient prior to the ceasefire.18 The results, reported in Online Appendix Table A.4, show no evidence of differential trends before the ceasefire for any of the nine educational outcomes. In addition, we perform a placebo exercise in which we estimate the main specification (Equation 1) but only for the pre-ceasefire period (2009–2014) and use as “placebo ceasefire” a series of dummies that equal one starting each year from 2010 to 2013. The results are reported in Online Appendix Tables A.5 and A.6 (for the schooling and the academic achievement outcomes, respectively).19 We find no significant changes in any of the placebo exercises and in any of the outcomes in municipalities highly exposed to FARC violence relative to other areas.
Lastly, we corroborate that other key municipal characteristics that likely affect educational outcomes do not change differentially in municipalities highly exposed to FARC violence neither before nor after the start of the ceasefire. To that end we estimate Equation 2 on outcomes such as municipal income, expenditures, and transfers from the central government; variables related to agricultural productivity; and health conditions and violence. We plot the estimated coefficients in Online Appendix Figure A.4. With the exception of guerrilla violence, which naturally increases differentially in treated municipalities before the ceasefire (consistent with our treatment definition) and decreases after the ceasefire (consistent with what the ceasefire achieved in terms of violence reduction), all the other potential confounders do not differentially change around the start of the ceasefire in treated municipalities.20
Taken together, these results illustrate the absence of differential pre-trends before the ceasefire, both in our outcome variables and in key potential confounders. This is key to validating our empirical strategy and providing credibility to our main result, namely that FARC’s permanent ceasefire triggered a short-term differential evolution of educational outcomes that facilitated human capital accumulation in exposed areas.
D. Further Robustness
We now assess the robustness of our main findings to a series of empirical exercises.
1. Measurement of exposure to FARC violence
Our results are largely robust to using two alternative measures of exposure to FARC violence. This is reported in Columns 1 and 2 of Tables 4, 5, and 6, respectively, for the schooling outcomes, the outcomes associated with academic achievement in the math tests, and those associated with the achievement in reading tests. In particular, we show that the results persist when using a more stringent measure of “high exposure” to FARC violence, namely an indicator that takes the value of one for municipalities above the median of the empirical distribution of per capita FARC attacks, conditional on experiencing at least one attack (Column 1). They are also robust to using the continuous treatment, namely per capita FARC attacks. In both cases, we find that, with few exceptions, our coefficients of interest remain significant, suggesting that our coding of the treatment variable does not drive our main results.
Robustness Exercises: Educational Outcomes
Robustness Exercises: Math Learning Outcomes
Robustness Exercises: Reading Learning Outcomes
2. Weights
Our main findings are based on estimating Equation 1, weighting the observations by the average municipal enrollment rates in the pre-ceasefire period (2009–2014). By doing so we increase the relative importance of municipalities that have a larger student population, thus taking into account the heterogeneity that Colombia exhibits in terms of the size of municipalities. As suggested by Solon, Haider, and Wooldridge (2015), we also report the unweighted version of the same specification in Column 3 of Tables 4, 5, and 6. The results are qualitatively unchanged, with the size of the estimated coefficients larger for dropout rates and somewhat smaller for the educational achievement outcomes. In some cases, however, they fall short of reaching statistical significance at conventional levels. This implies that the effect that the ceasefire has on educational outcomes is relatively larger in medium to large municipalities.21 It is also worth noting that the choice of the weight does not drive our main results. Online Appendix Tables A.9 and A.10 show that our findings are unchanged when we use alternative weights, namely total enrollment in 2014, year-by-year enrollment, and total municipal population in 2014.
3. Municipality controls
Recall that our preferred specification includes municipality characteristics before the ceasefire interacted with the post-ceasefire period dummy. Column 5 of Tables 4, 5, and 6 reports a version of this regression in which, following Belloni, Chernozhukov, and Hansen (2014), the controls are selected using machine learning. In this way we are agnostic about which municipality characteristics are ex ante more related to educational outcomes and exposure to FARC violence.22 Our results are robust to this exercise.
Alternatively, we estimate a propensity score for the indicator of being highly exposed to FARC violence and add it to the main specification as the only municipal characteristic interacted with the post-ceasefire period dummy.23 This captures differential changes in educational outcomes parametrized by a statistic that summarizes several pre-ceasefire observable characteristics that are potentially related to the exposure to FARC violence. Column 6 of Tables 4, 5, and 6 shows that our results are robust to this control.
4. Comparison municipalities
One threat to our identification strategy is that municipalities exposed to FARC violence were different from areas not exposed along with other (nonobserved) characteristics and that in 2014 there might have been some shock (other than the ceasefire) that differentially affected these municipalities because of such characteristics, but not because of the prior exposure to FARC violence. To alleviate this concern we estimate our main model using different control municipalities, which we select through matching procedures. First, we keep in our sample only municipalities that have been exposed to violence from FARC or other armed groups between 2009 and 2014.24 In this way, we are keeping constant the exposure to conflict-related violence. The main assumption behind this strategy is that in the absence of the ceasefire, the trends in educational outcomes in municipalities previously affected by FARC violence would have been the same as in municipalities affected by violence perpetrated by other armed groups. Column 8 of Tables 4, 5, and 6 shows that our estimates are robust to using these municipalities as the control group.
Alternatively, following Crump et al. (2009), based on the estimated propensity score described in Section V.D.3, we truncate the sample to increase the overlap of treated and control municipalities in terms of various municipality characteristics. We perform this truncation using the optimal cutoff suggested by Crump et al. (2009), which in our case is 3.8 percent. Column 9 of Tables 4, 5 and 6 shows that our results are also robust to this sample truncation strategy.
5. Placebo simulation
We also conduct a permutation test by randomly assigning an indicator of pre-ceasefire exposure to FARC violence to all municipalities over many iterations. This test provides us with a distribution-free estimate of the probability that our coefficient of interest arises by chance. In particular, we randomize the assignment at the country level in a way that is consistent with the observed distribution of municipalities exposed to FARC violence. Online Appendix Figure A.5 reports results. Reassuringly, our estimated coefficients (red vertical line) are above the 96th percentile of the resulting distributions.
6. Influential observations
A final concern is that a few outliers might drive our results. Online Appendix Tables A.7 and A.8 show that our results are robust to dropping influential observations using three different winsorization cutoffs, thus alleviating this potential concern. Alternatively, we also check whether our main results are driven by a particular treated municipality or by one specific department.25 To that end we remove treated municipalities (see Online Appendix Figure A.6) or entire departments (see Online Appendix Figure A.7) one by one and reestimate our baseline results. In general, all coefficients remain stable and statistically significant.
VI. Mechanisms
In this section, we explore the empirical relevance of several potential mechanisms through which the absence of violent conflict can affect educational outcomes and promote human capital accumulation in municipalities previously affected by FARC violence. We explore the role of the overall victimization of civilians, the recruitment of child soldiers, the availability of peace building and recovery public investments, and potential countervailing heterogeneous effects generated by the existence of legal and illegal economic opportunities in treated municipalities.
A. Victimization
Households affected by violence are likely to remove their kids from school because of uncertainty and perceptions of fear (Justino 2016). This is potentially quite relevant in our context, as the Colombian conflict resulted in around 8.8 million officially recognized victims, equivalent to 17 percent of the country’s population.26 Victimization events include forced displacement, killings, and kidnappings. Schools’ infrastructure and resources were not exempt from violence during the course of the conflict (CNMH 2017). School facilities were used as supply centers and camping areas by both illegal armed groups and state security forces. Moreover, both students and teachers faced threats and indoctrination attempts and were exposed to various forms of attacks, often being caught in crossfire. Records from the Colombian teachers’ union (FECODE, by its Spanish acronym) show that during the period 2009–2014, there were on average 22 homicides of teachers per year. Instead, after the start of the ceasefire (during 2015–2016), when FARC’s bellicose activity dropped by 98 percent, there were eight teacher homicides per year on average (Fecode 2016).
Overall, students and teachers attending and working in schools located in conflict-affected areas faced a nonnegligible risk of victimization. We assess whether and to what extent our main results are driven by the large reduction of victimization following the ceasefire. To that end, we estimate Equation 3 to explore if there are any heterogeneous effects in municipalities that, prior to the ceasefire, suffered particularly high levels of violence. We do so by looking at three different but complementary sources of variation that capture pre-ceasefire levels of violence. First, we consider the violence perpetrated by other armed groups (in addition to FARC), hence identifying areas that were more contested and therefore experienced more violence.27 Second, we leverage on the episodes of explosion of land mines, which represents one of the main strategies of inflicting fear on the civilian population during the conflict and that plummeted after the ceasefire by 67 percent in municipalities exposed to FARC.28 Finally, we explore heterogeneous effects based on how exposed the municipality was to forced displacement. In municipalities exposed to FARC violence, forced displacement dropped by 44 percent after the ceasefire (see Table 1).29
The results from these tests are reported in Columns 1–3 of Table 7 for the case of schooling outcomes and in Columns 1–3 and 6–8 of Table 8 for the case of learning outcomes associated with math and reading test scores, respectively. We find that the average reduction in dropout and grade failure rates is exacerbated in FARC-exposed municipalities that were disputed by other illegal armed groups (Panels A and B, Column 1 of Table 7), and so are the math test scores in Grades 3 and 5 (Panels A and B, Column 1 of Table 8) and the reading test scores in Grade 3 (Panels A, Column 6). Pre-ceasefire land-mine victimization also exacerbates the average reduction in dropout and failure rates (Panels A and B, Column 2 of Table 7), as well as the average increase of math test scores in Grade 5 (Panel B, Column 2 of Table 8) and that of reading test scores in Grade 9 (Panel C, Column 7). Finally, the intensity of pre-ceasefire forced displacement amplifies the reduction in dropout rates and the increase in the number of high school graduates (Panels B and C, Column 3 of Table 7).30
Municipal Heterogeneous Effects: Schooling Outcomes
Municipal Heterogeneous Effects: Learning Outcomes
Taken together, this evidence points to the conclusion that the generalized collapse of violence that followed the permanent ceasefire declared in the midst of peace negotiations between FARC and the Colombian government is an important driver of the post-ceasefire positive evolution of a large set of educational outcomes.
B. Recruitment of Child Soldiers
According to official records, FARC is responsible for 54 percent of the 16,879 identified cases of illegal child recruitment into armed groups in Colombia between 1960 and 2016. Most of these children were boys (68 percent) and were recruited when they were 12–16 years old (CNMH 2017).31 Clearly, to the extent that the recruitment of children stops after the de jure or de facto end of conflict, child recruitment during conflict is an obvious candidate for explaining our findings.
Our test of whether this is the case is twofold. First, we estimate our main empirical specification (Equation 1) using as the dependent variable the number of children recruited by any armed group as well as the number of cases of recruitment. These data are recorded by the Centro Nacional de Memoria Histórica (Colombia’s Truth Commission) and are available at the municipality–year level.32 Second, we estimate Equation 3 to explore potential heterogeneous effects across key student or school characteristics that are likely correlated with the recruitment of child soldiers. In particular, we are interested in learning whether the reduction of school dropout is higher for boys, for kids in the age window that is associated with the highest incidence of abduction of children, or for rural or public schools, which are more likely to be located in areas where the recruitment of children took place (relative to schools located in urban areas or to private schools).
Table 9 reports the results of the first test. We find a significant reduction in recruitment equivalent to 1.3 children (and 1.2 cases) after the ceasefire in municipalities affected by FARC violence. This reduction represents 6 percent of the average and 0.24 of the standard deviation of all the recruitment that took place during our pre-ceasefire sample period. In order to get a sense of the empirical relevance of these estimates, we perform a bounding exercise that is specific to one of the outcomes under study, namely dropout rates. In particular, we take the year of the highest recruitment prior to the ceasefire and divide it by the number of municipalities exposed to FARC violence over this period. This yields an average of three recruited children per affected municipality. If we assume that these three children stayed in school after the start of the ceasefire, then this could explain up to 8.5 percent of the average estimated reduction in the school dropout rate.33
Effects on Recruitment
The results of the second test are reported in Table 10. We find that the reduction in dropout and failure rates is exacerbated in primary school grades, with children aged six to 11 (Panels A and B, Column 1). This result goes against child recruitment as a potential mechanism, insofar as the majority of abducted children by illegal armed groups in Colombia are in the age bracket of 12–16.34 Importantly, we do not find any other robust heterogeneous effect in terms of the individual and school-level characteristics potentially associated with child soldiering.
Student and School Heterogeneous Effects
Taken together, these results suggest that, even if child soldiering is a phenomenon of foremost importance in the Colombian conflict, other mechanisms are likely more empirically relevant to explain the evolution of educational outcomes after the ceasefire in areas traditionally exposed to FARC violence.
C. Peace and Recovery Investments
The effect of the ceasefire and the subsequent peace agreement on educational outcomes could at least partially be explained by government efforts to consolidate peace and help recover conflict-affected areas through various investments in infrastructure and the creation and expansion of social programs. Several pieces of evidence suggest that the empirical relevance of this potential mechanism is at best weak. First, the timing of most of the positive effects that we unveil on educational outcomes is not completely consistent with that of the post-conflict government investment efforts. Indeed, as reported in Figure 4, the decrease in dropout and grade failure rates, as well as the increase in reading and math test scores starts, by and large, in 2015. That is, the short-term generalized improvement in schooling and academic achievement is evident right after the start of the ceasefire, but before the start of the implementation of the peace agreement and hence before the rollout of most of the public peace building initiatives (2017 onwards).
Second, Online Appendix Figure A.4 shows that key variables related to the municipal capacity of investing in peace building and post-war recovery, namely total municipal income and expenditure (Panels B and C, respectively), and total transfers along with those received specifically from the central government (Panels D and E) do not differentially increase in treated municipalities during the post-ceasefire period. Moreover, our estimated coefficients are unchanged if we control for an indicator that takes the value one after the signing of the peace agreement (2017 onwards) for the 170 municipalities prioritized by the agreement to receive peace building and recovery investments (see Online Appendix Table A.11).
Third, and related to the previous point, our findings do not seem to be explained by the rollout of social programs or subsidies. To show this, we mapped all the antipoverty and social programs that were active at the time of the ceasefire and studied each one of them to check which actually changed in the post-ceasefire period (for instance, in terms of the number of beneficiaries or the type or the amount of the subsidy provided). We also checked whether any new social program was established during the post-ceasefire period or any of the preexisting programs disappeared. To that end, we studied the official documentation and operational annexes of the programs, as well as the official resolutions that rule them. From the universe of social programs analyzed, we ended up selecting two social programs that faced some change: Familias en su Tierra and IRACA.35 The first program helps former victims of forced displacement to return and settle in their land. The second attends ethnic households that have been victims of forced displacement. In Column 7 of Tables 4, 5 and 6 we control for the per capita number of beneficiary households of each of these two programs, averaged over the years for which lists of beneficiaries were available, interacted with the year fixed effects. We observe no changes in our coefficients of interest.
Finally, in Online Appendix Table A.12, we also rule out that the post-ceasefire improvement of educational outcomes in municipalities affected by FARC violence is explained by a differential change in school conditions or the supply of schooling in such areas. Specifically, we do not find any significant differential increase in the student-to-teacher ratio (Column 1) or the number of teachers (Column 2). We also do not find any differential change in the opening of new schools (Columns 3 and 4).36 This is, however, not completely surprising given the short-term span of the analysis.
D. Economic Opportunities
Economic opportunities may attenuate or exacerbate the effect of the ceasefire on educational outcomes, depending on the extent to which they are substitutes or complements of schooling and human capital accumulation. These countervailing effects may be present, for instance, when comparing illegal economic activities that are intensive in unskilled (and often underaged) labor—like growing illicit crops—to legal economic activities that increase the demand for skilled labor. While the former would increase the opportunity cost of schooling (offsetting our findings), the latter would increase its returns (strengthening them). Indeed, the depression of economic prospects during a conflict is usually considered a key driver of the documented negative effect of conflict on human capital accumulation (see Justino 2016 for a recent review of this literature).
To test this conjecture, we first estimate Equation 3 to look at the differential change in educational outcomes after the ceasefire in municipalities with high suitability to grow coca, cocaine’s main precursor. As shown by Prem, Vargas, and Mejía (2021), in these areas, coca cultivation spiked after the ceasefire because of a government announcement that material incentives would be given to coca growers in order to substitute the illegal crop for legal alternatives. Indeed, because of its unusually large profitability, coca production is likely to attract child labor (Angrist and Kngler 2008). The results are reported in Columns 4 and 5 of Table 7 for the case of schooling outcomes and in Columns 4 and 5 (9 and 10) of Table 8 for the case of the learning outcomes associated with math (reading) test scores.37 Consistent with the conjecture that the opportunity cost of schooling increases in places with more coca cultivation, we find that the decrease in dropout and grade failure rates is attenuated in municipalities with more exposure to coca production (Panels A and B, Columns 4 and 5 of Table 7). Moreover, as shown in Panels C and D of Online Appendix Figure A.8, the size of the attenuation is somewhat larger for age groups that are more likely to substitute schooling for this type of work, especially for the case of failure rates (Panel D). Finally, the increase in math test scores in Grades 5 and 9 is also partially offset (Panels B and C, Columns 4 and 5 of Table 8), along with the increase in reading test scores in Grade 5 (Panel B, Columns 9 and 10 of Table 8).
Second, using firm-level entry data, Panel I of Online Appendix Figure A.4 shows that during the first two years after the start of the ceasefire, the optimism generated by the large decrease in violence in formerly FARC-affected municipalities was accompanied by a differential establishment of new firms in these municipalities.38 To the extent that this extensive margin form of differentially larger formal economic activity increases the expected returns of human capital accumulation, then we should expect this to be a relevant mechanism for our results. This is indeed what we find in Column 6 of Table 7 for the case of schooling outcomes and in Column 6 and 12 of Table 8 for the case of the learning outcomes associated with math and reading test scores, respectively. The positive effects of the ceasefire on all educational outcomes (except high school graduation and the performance on Grade 9 reading test scores) are strengthened in municipalities that experienced a larger increase in firm entry in the two years after the ceasefire.39
VII. Conclusion
We study the short-term effect of Colombia’s recent efforts to bring the conflict with the FARC insurgency to an end on a range of educational outcomes that lie at the core of human capital accumulation. Our findings show that the permanent ceasefire declared by FARC during peace negotiations with the government triggered a large differential reduction in school dropout and grade failure rates, as well as an increase in high school graduation and test scores in reading and math, in the areas most affected by FARC violence prior to the ceasefire relative to other areas. Specifically, we find that municipalities exposed to FARC violence prior to the ceasefire experience after the ceasefire a differential 19 percent reduction in dropout rates, 13 percent reduction in failure rates, 19 percent increase in high school graduation, and about 1.5 percent increase in standardized test scores at different school grades.
We rule out that this effect is entirely driven by the wartime recruitment of child soldiers or by peace building and post-war recovery efforts and posit that, instead, the post-ceasefire generalized reduction in victimization rates and the availability of new economic opportunities that increase the expected returns to schooling are two potentially important mechanisms. In contrast, we also find that the generalized improvement in schooling and academic achievement is attenuated in places more prone to host illegal economic activities intensive in unskilled labor and likely to use child labor.
The findings are important for the design of policies that seek to maximize the long-term peace dividend of the end of a five-decade-long conflict. Indeed, the short-term improvement of a variety of educational outcomes is a necessary condition to counteract the long-term human capital costs of civil war, which has been widely documented for many countries and contexts, including Colombia. The importance of it cannot be emphasized enough. According to UNESCO (2011), while 65 percent of primary-school-age children attend school in conflict-affected countries, the same figure for comparable low-income countries is 86 percent.
In the specific case of Colombia, our findings highlight the importance of the ongoing civil society efforts to consolidate a resilient peace. The peace agreement with FARC was signed in September of 2016 but got rejected by a 0.5 percent vote margin in a referendum that took place in October that year. In 2018, the party that promoted the “no” vote rose to power, and the implementation of the agreement (which was endorsed by Congress in December 2016 after the negotiating team made some adjustments following the electoral defeat in the referendum) has slowed down significantly since then. The short-term human capital peace dividend highlighted here points to the importance of continuing with all efforts towards the implementation of the peace agreement. We encourage the government and other stakeholders in Colombia to do so.
Footnotes
The authors thank Felipe Barrera, Raquel Bernal, Leonardo Bonilla, Mathieu Couttenier, Antonio Hernández, Charu Prem, Jacob Shapiro, Oliver Vanden Eynde, Alexander Villarraga, Austin Wright, and seminar participants at Banco de la República (Bogotá and Medellín), LACEA, Rosario-Andes Taller Applied (RATA), and the 2019 PSE Workshop on Conflict for helpful comments and suggestions. Nathalie Basto, Carolina Bernal, Gabriel Suárez, Andrés Rivera, and Catalina Zambrano provided excellent research assistance. Namen participated in this project during a postdoctoral fellowship at Universidad del Rosario. The views expressed in the paper do not necessarily reflect those of Innovations for Poverty Action. The research reported in this paper was made possible in part by a grant from the Spencer Foundation (#202000039) as well as by the Colombia Científica-Alianza EFI Research Program 60185 with contract No. FP44842-220-2018. The views expressed are those of the authors and do not necessarily reflect the views of the Spencer Foundation. The authors have nothing to disclose. The data used in this article are available online at ICPSR (https://doi.org/10.3886/E135341V1).
Color versions of some graphs in this article are available through online subscription at: http://jhr.uwpress.org
Supplementary materials are freely available online at: http://uwpress.wisc.edu/journals/journals/jhr-supplementary.html
↵1. Additional long-term consequences of the effect of conflict on human capital include the delayed transition from agriculture to manufacturing due to the shrinkage of sectors that specialize in human capital intensive labor (Fergusson, Ibañez, and Riaño 2019).
↵2. The ceasefire was declared on December 20, 2014 and constituted one of the main milestones of the four-year-long peace process that started in October 2012. Because it was largely met, it signaled a credible commitment by FARC to lay down their weapons. On August 29, 2016, FARC and the government reached a definitive bilateral ceasefire, and later that year, the final peace agreement was signed by both parties, putting an end to a five-decade-long conflict.
↵3. See, for instance, Akbulut-Yuksel (2014); Akresh and de Walque (2011); Chamarbagwala and Morán (2011); Leon (2012); Shemyakina (2011); Justino (2012); Brück, Di Maio, and Miaari (2019); and Bertoni et al. (2019).
↵4. For evidence of the effect of conflict on child labor, see Akresh and de Walque (2011), Rodríguez and Sánchez (2012), and Di Maio and Nandi (2013). The effect of conflict on health outcomes is studied by Ichino and Winter-Ebmer (2004); Shemyakina (2011); Bundervoet, Verwimp, and Akresh (2009); Agüero and Deolalikar (2012); Parlow (2012); Camacho (2008); and Valente (2014) among others. Child soldiering has received less attention in the literature, probably due to the lack of reliable data. One exception is Blattman and Annan (2010).
↵5. We also show that our results are not driven by any specific department or treated municipality. Moreover, we conduct a permutation test by randomly assigning the FARC violence treatment many times, thereby ruling out that our results arise by chance.
↵6. FARC’s offensive activity dropped by 98 percent after the ceasefire (CERAC 2016), and the number of land mines explosions dropped by 76 percent.
↵7. Source: Victims’ Registry, from the Unit for the Victims Assistance and Reparation, November 2020 figure (https://www.unidadvictimas.gov.co/en, accessed November 9, 2022).
↵8. Santos was ultimately awarded the Nobel Peace Prize in 2016 “for his resolute efforts to bring the country’s more than 50-year-long civil war to an end.”
↵9. Our results are robust to using all the country’s municipalities.
↵10. Note that our measure of dropout does not include students that leave a school after the end of the academic year. Also, because we do not have individual-level data for students with identifiers, we cannot follow their trajectories to assess whether they return to the school at some point in the future.
↵11. We do not include test scores from the high school national test (“Saber 11”), as these are not comparable across years during our sample period, due to several methodological changes documented in ICFES (2019).
↵12. Noche y Niebla sources include “1. Press articles from more than 20 daily newspapers of both national and regional coverage. 2. Reports gathered directly by members of human rights NGOs and other organizations on the ground such as local public ombudsmen and, particularly, the clergy” (Restrepo, Spagat, and Vargas 2004, p. 404). Notably, since the Catholic Church is present in even the most remote areas of Colombia, we have extensive coverage of violent events across the entire country.
↵13. The 2011–2014 period covers the years after president Juan Manuel Santos took office and before the beginning of the permanent ceasefire.
↵14. As a robustness check, we estimate our model using a variance–covariance matrix that takes into account cross-sectional dependence in the error term following Conley (1999, 2016).
↵15. As an additional exercise to assess the robustness of our standard errors, we follow Bertrand, Duflo, and Mullainathan (2004) and collapse our data before and after the ceasefire to deal with potential serial correlation. Column 4 of Tables 4, 5, and 6 reports these results, which reassure the validity of the baseline estimates of Table 2.
↵16. To this end, we use the correction proposed by Jones, Molitor, and Reif (2019).
↵17. For conciseness, all the robustness exercises, as well as the analysis of the mechanisms, focus on this demanding specification.
↵18. The specification we run is
, where Trendt is a linear trend and we restrict the sample to the years 2009–2014. Our parameter of interest, β, shows whether there are differential trends between municipalities exposed to FARC and not exposed.↵19. Note that in the case of the academic achievement outcomes, there is only one placebo ceasefire year (2013) because of the reduced sample period for which these outcomes are available (see Section III).
↵20. This excludes the variable that captures the differential creation of new firms in treated municipalities (Panel I). New firms are created after the ceasefire disproportionally more in places traditionally highly exposed to FARC’s violence. This is indeed consistent with our findings, and we discuss this potential mechanism more in detail in Section VI.
↵21. The average (median) municipal size, out of the almost 1,100 Colombian municipalities, is 21,000 (33,000) people.
↵22. The set of potential controls includes the logarithm of population, the size of the municipality, the distance to department capital, the share of rural population, a poverty index, the municipal tax revenue per capita, the municipal fiscal deficit, a coca suitability measure estimated by Mejía and Restrepo (2015), the average elevation of the municipality, the area of the municipality, and the illiteracy rate. We select the controls for each dependent variable separately.
↵23. To construct the propensity score, we used the same set of controls mentioned on Footnote 22.
↵24. Exposure here is defined in the extensive margin, so these are municipalities that suffered at least one attack by any illegal armed group during 2009–2014.
↵25. Colombia has 32 departments, equivalent to U.S. states.
↵26. See the Victims’ Registry of the Colombian Unit for the Attention and Reparation of Victims (https://cifras.unidadvictimas.gov.co/Cifras/, accessed November 14, 2022).
↵27. Following Prem et al. (2020) and Acemoglu, Garcia-Jimeno, and Robinson (2015), we use a measure of exposure to other armed groups that is a weighted average of exposure to other groups based on the distance to municipalities with presence of other armed groups. This measure takes into account the ruggedness and other physical characteristics of the terrain that affect intermunicipal connectivity. The data on other armed groups comes from the same source as that of FARC’s attacks (see Section III).
↵28. This is explained both by the reduction in the placement of new mines and by a rapid nation-wide response in demining efforts. The latter owes to the humanitarian demining agreement, signed by FARC and the government in March 2015—well before signing the final peace agreement in the second half of 2016. Shortly after the demining agreement, FARC started providing intel on the location of unexploded mines across the country.
↵29. Forced displacement statistics per municipality and year come from Colombia’s Victim’s Unit, and can be found at https://cifras.unidadvictimas.gov.co/Cifras/ (accessed November 14, 2022).
↵30. However, it also attenuates math test score in Grade 3 (Panel A, Column 3 of Table 8).
↵31. These patterns are consistent with those found for the case of Uganda’s Lord’s Resistance Army (Blattman and Annan 2010; Beber and Blattman 2013).
↵32. Unfortunately, there is no longitudinal data on child recruitment that distinguishes the group responsible for the abduction.
↵33. To see why, recall that the estimated average dropout reduction is 19 percent (see Column 2 of Table 2). Also, the average enrollment in municipalities affected by FARC violence is 186.8. Now, 3/(0.19 × 186.8)≈0.085.
↵34. Further, in Online Appendix Figure A.8, we examine heterogeneous effects by school grade level. Each coefficient corresponds to a separate regression using as the dependent variable grade-level dropout rates (Panel A) and failure rates (Panel B). The estimates show that the effects are similar for younger and older children, and if anything slightly smaller (in absolute value) for the latter.
↵35. The other three, which did not present any change that was contemporary to the ceasefire, are Colombia Mayor, Jóvenes en Acción, and Familias en Acción.
↵36. We define school opening as the number of schools that appeared in a municipality during a particular year and that were not in the database the year before. The information on the school identifiers is taken from the yearly school census (Form C-600) described in Section III.
↵37. Odd columns use a continuous version of the coca suitability index (computed by Mejía and Restrepo 2015), and even columns use a discrete (dichotomic) version that takes the value one for municipalities above the median of the empirical coca suitability distribution.
↵38. This short-lasting effect is probably associated with the fact that, upon the signature of the final peace agreement at the end of 2016, its implementation was largely slowed down by the lack of Congress support to President Juan Manuel Santos (2010–2018) and then by the lack of political willingness of the administration of President Iván Duque (2018–2022).
↵39. A caveat of this estimation is that because differential firm creation is measured by comparing post-ceasefire years with the pre-ceasefire averages, this triple interaction constitutes a “bad control” (Angrist and Pischke 2008). Therefore, these results should be interpreted with caution.
- Received March 2020.
- Accepted March 2021.
