Persistence of the armed conflict in Colombia: an agreement without peace Open Access

The Colombian armed conflict has its origins in the 1930s with the occupation of land by peasants and the subsequent appearance of organized groups that sought to legitimize their possession through the production and productivity of the land, as well as to defend themselves from the actions of the Army that sought to evict them (Trejos, 2008). The subsequent “bipartisan” ^[1] violence led to organizational processes in the communities to defend themselves from liberal or conservative actions, and on the other hand, it meant several pacifying operations by the army as in the case of President Uribe's democratic security program, led to the emergence of new insurgent and paramilitary groups that defended the status quo (Sierra, 2019; Trejos, 2013).

The recognition of this armed conflict implied heated debates with governments in power that did not want to acknowledge the existence of armed combats in the territory. It involves the state, insurgent groups known as guerrillas, self-defense groups, paramilitaries and criminal gangs. In addition to these actors, the existence of a certain number of victims and density of violent acts, allow us to catalog an internal armed conflict, (Henderson & Singer, 2000; Leongómez, 2015; Uprimny, 2005).

The complexity of the Colombian armed conflict encompasses political, economic, social, institutional, and environmental dimensions. This multifaceted nature is reflected in a wide array of explanatory studies that, over time, have provided insights into the causes and consequences of the conflict, the sustainability of post-conflict peace, and, more recently, the factors driving the emergence and resurgence of internal armed conflicts in Colombia and civil wars in other countries.

In this regard, it is clear that the variables that trigger conflicts are not the same as those related to perpetuation (Yaffe, 2011). Although many investigations coincide with the statement that the conflict in Colombia emerged due to poverty, poor living conditions and absence of the state, it is not the same when it comes to perpetuation factors (Gómez, 2001; Sánchez & Chacón, 2005; Valenzuela, 2002).

In this sense, it is important to research the factors that determine the persistence of armed conflict in Colombia. This study proposes as a novelty to incorporate the viability hypothesis with some factors related to the grievance that explain the persistence of the armed conflict as a consequence of decreasing recruitment costs for the insurgent groups, a situation that contributes to the opportunity structure and financial viability of the conflict.

While inequality in land distribution is a factor that generates armed conflict, it does not behave in the same way to perpetuate the conflict. In the latter case, the conflict actors dispute territorial control to gain access to the exploitation of natural resources and therefore the explanatory factor here is the greed factor and not the economic factor of inequality that originated the conflict (Gutiérrez & Vargas, 2016). The signing of the Peace Agreement in 2016 by Revolutionary Armed Forces of Colombia (FARC), the oldest guerrilla group of the region, left the country in a post-agreement scenario. That was marked by focused violence, territorial disputes, homicides, forced displacements, extortions, vacunas ^[2] and robberies, which revealed the need to research thoroughly into those factors that prolong the conflict in Colombia and that the peace process has not been able to remedy.

The emergence, intensity and duration of civil wars or internal armed conflicts are usually explained by the thesis on greed due to legal and illegal lootable natural resources, and their incidence is greatly accepted, as it is one of the most widely used study approaches in the political economy of armed conflicts. Studies by Collier and Hoeffler (2004) indicated that the dependence of the state on the export of primary products leads to a greater risk of civil war than those states that are poor in natural resources. The same occurs with the large reserves of oil, precious stones and narcotics, which exacerbate and extend conflicts (Collier et al., 2003; Fearon & Laitin, 2003; Ross, 2006), while lootability and legality of resources affect the effectiveness of policies aimed at ending conflicts (Le Billon, 2009).

Likewise, the ethnic fragmentation associated with the concept of polarization can be a significant variable in civil wars and conflicts (Garcia-Montalvo, Reynal-Querol, Reynal-Querol, & Polarization, 2004) and the effect that exposure to armed conflict has on children's schooling (Rodriguez & Torres, 2009) has an impact on conflict, in particular, the latter, which has an impact on recruitment and the perpetuation of conflict.

Although drugs and illicit crops are not the initial cause of conflict, they expand and exacerbate preexisting conflicts by promoting cooperation between antagonistic actors (Ross, 2004). It is important to highlight the relevance of institutional and political factors to explain violence. The low political participation, poor administration of natural resource revenues and the absence of the state are conducive to the emergence of armed conflicts (Bates, 2008).

In the Colombian armed conflict, the so-called objective or grievance factors, such as poverty and inequality, lose explanatory importance in terms of intensity and territorial expansion (Gaitán, 2001), despite the evidence that correlates the decrease in coffee prices with the intensification of the conflict. Social programs to mitigate poverty and unemployment that have a moderating effect on violence (Dube & Vargas, 2006) explain this situation.

Drug trafficking in Colombia is undoubtedly a perpetuating element of the armed conflict since it provides the financial reason and support to criminal gangs. They are heirs of the self-defense groups, with all the legacy of violence and economic greed by participating in looting and legal extraction, which confirms that lootable resources have an impact on the perpetuation of the conflict (Centro Nacional de Memoria Histórica (CNMH), 2016; Echandía, 2013; Restrepo, Spagat, & Vargas, 2006). As Fernández (2014) concludes, the emergence of criminal gangs does not constitute a new actor in the conflict, but the reappearance of an old one because of the State's action to combat it. Recently, the lootability of gold gains ground as a source of financing for insurgent armed groups and evidence indicates that it is an explanatory factor for the intensity and extension of the armed conflict (Escobedo & Guío, 2015; Giraldo, 2013; Massé & Camargo, 2012; Ortiz-Riomalo & Rettberg, 2018).

The armed conflict manages to expand independently of the presence of the state in those regions with significant natural resources and great economic activity (Bottia, 2003; Salazar, 2005), which makes it clear that the lootable resources factor outweighs the institutional factor. Thus, the main independence and fiscal strength of local governments is positively correlated with the intensification of armed violence that seeks to appropriate public goods and resources and influence the political and electoral results at the irregular groups’ convenience (Sánchez & Chacón, 2005).

Thus, institutional weakness and corruption, together with legal and illegal actors within the political system, made the spread of violent acts in the national territory possible (Romero & Arias, 2008). This situation perpetuates the conflict (Gutiérrez, 2015) and destabilizes the state with weak bureaucratic institutions that deteriorate democracy by introducing candidates aligned with criminal agents, such as drug traffickers, paramilitaries or guerrillas (Ávila & Velasco, 2012).

Numerous studies aim to understand the persistence of the armed conflict in Colombia even after the signing of the Peace Agreement with FARC but they fail to answer the questions posed in this research regarding the main factors that determine the perpetuation of this armed conflict.

The identification of the factors in this research allowed us to propose a model that estimates the probability of recurrence of the armed conflict taking into account 23 states during 2008–2018. This period includes two years after the signing of the Agreement (2016), which enabled the prediction of the most exposed states to suffer an unfading conflict. Unfortunately, due to the scarce information in several variables, we cannot analyze a longer period.

The used methodology estimated nonparametric survival and risk functions and implemented a parametric probit model for panel data, inspired by the feasibility hypothesis of Collier, Hoeffler and Rohner (2009), the highlight the practicalities of the rebellion as is the case of the availability of resources, recruits and military capacity. This perspective shifts the explanatory focus from the why of the rebellion to showing its viability. We complemented this with some variables that respond to the greed and grievance thesis of Collier and Hoeffler (2004) for their indirect contribution in decreasing the costs of rebellion and therefore, in making the conflict viable.

The results obtained were examined based on transdisciplinary perspectives that corroborated that the viability hypothesis has explanatory capacity in the continuity of the armed conflict. Likewise, it was found that political fragmentation contributes to rebellion, and that variables representing the dimensions of grievance and institutional presence contribute to the opportunity structure that makes the conflict in Colombia militarily and economically viable.

Secondary information is obtained from different sources including the National Administrative Department of Statistics (Departamento Administrativo Nacional de Estadística, DANE), the Unitary Victims’ Registry (Registro Único de Víctimas, RUV), the United Nations Office on Drugs and Crime (UNODC), among others. The following is a descriptive analysis of the relevant variables in the analysis ^[3]. Table 1 shows the average and the number of observations at country level, and by region depending on its level of development ^[4].

The variable homicides (hom) is a dichotomous one. Therefore, if the number of homicides in a department (i) exceeds the average number of homicides in the sample (23 departments) in year (t), then it is equal to 1 and 0 otherwise. It is worthing mentioning as in any dichotomous variable, the sample average coincides with the possibility of success. Therefore, as the mean of the variable hom is equal to 30%, it means that this department exceeds the average of the 23 departments by 30 homicides. However, in regard to regions, the situation changes clearly since the most developed territories show a very high rate of conflict-related homicides (close to 61%) while in the least favored territories it is lower (21%). The possible explanation is that the high income or economic activity in these regions attracts Nonstate Armed Actors, hereinafter NSAA, to entrench the conflict.

On the other hand, the rural population rate shows an expected result since it is higher in the less favored regions and conversely lower in the more advanced ones. It is expected that in the regions with high rural rates violence is more entrenched than in the others. About the number of political parties, there is a similar behavior at an aggregate level and per region, since it is around 50%, which indicates that fragmentation and multiparty electoral competition are very similar throughout the national territory.

The Gini and poverty numbers are related since income inequality is very high in poorest regions and low in richest regions. Specifically, at macro-level, the Gini coefficient is close to 50% on average; in the least-developed regions, it surpasses this figure, whereas in the more developed regions, it falls below it. This behavior is in line with poverty, which is very similar for the whole territory and for the least advanced regions (15.2%), but it is almost half in the rest of the country (7.2%).

The exploitation of mineral resources such as gold and coal shows similar in the country. In addition, the presence of deposits is close to 60 and 40%, respectively in the least-developed region. Nevertheless, in the most advanced territories, although gold deposits are also similar to the rest of the country (66%), the three departments that make up the region have coal deposits. These characteristics are possibly very attractive to the NSAA because mineral resources can be used for their own benefit.

GDP per capita shows an expected behavior, the rate is higher in the most advanced regions and lower in the remaining ones. This is so because the departments of Cundinamarca (including the capital, Bogota), Antioquia (including its capital, Medellin) and Valle del Cauca (including its capital, Cali) generate more than 50% of the country's level of economic activity and concentrate 46% of the total population (Gomez, Fajardo-Hoyos, & Sarmiento-Castillo, 2022). On the other hand, primary education coverage is high and similar in the country (almost close to 90%). However, secondary education coverage, as we expected, is higher in the most advanced regions (44 vs 38%). Higher education (not shown in Table 1) reveals a similar pattern. We consider that the least educated regions are those where the conflict has a high possibility of consolidating over time.

Unemployment is similar throughout the territory regardless of the region (around 10%). The coca cultivation is higher in the most economically backward regions, which, we suspect, will induce the conflict to remain longer in these territories. Oil royalties are on the same path, since the most prosperous regions do not have this natural resource and their economies are based on services and industry largely.

In summary, we can arrive at these conclusions. First, the general behavior of the analyzed variables in Colombia generally coincides with the pattern shown by the same variables in the least favored regions. This justifies the territorial disaggregation in the analysis, since not doing so would hide what happens in the most advanced regions. Second, both regions have their own characteristics that could perpetuate the conflict, such as natural resources (gold, coal or oil) in the least favored regions and high levels of wealth in the most prosperous regions that are very attractive to violent groups in pursuit of economic profits.

This section is based on Llopis, Castillejo, and Sánchez (2022). It shows a nonparametric analysis that determines the duration of the armed conflict in Colombia over time. Specifically, Kaplan–Meier (KM) survival function are used to show the percentage of duration of the conflict after a given number of years. A proportional hazard model was also used where the duration of the conflict is treated as discrete time because the data, since they are annual, are recorded by intervals ^[5].

Thus, the duration is measured according to the seasons where there is an uninterrupted number of war-related homicides. We considered “uninterrupted” the period that begins in t with reported homicides but without any report in the previous period t–1 and ends when there are no victims in a subsequent period (Llopis et al., 2022). In other words, the persistence of the conflict is proxied by the seasons where violence does not stop.

In our case, the KM survival graph shows in the y-axis the persistence/continuity likelihood of the armed conflict during the time analyzed. That is, it shows the proportion of states/regions that continue to be violent at a specific moment in relation to the total number of states at risk after the problem has been identified. As time progresses, the conflict surviving likelihood falls, which will be reflected in a downward curve on the graph of the survival function. Figure 1 shows the KM function.

The results show that the survival function of the less-developed regions is always higher than the survival function of the more advanced regions, which indicates that the seasons of violence are more persistent in regions with greater social and economic problems. This may be due to the social, economic and political characteristics that make it easier to perpetuate the conflict. These facts will be analyzed below. Moreover, it is also observed that the deterioration of the conflict is faster in the first five years in the more developed regions while it falls more slowly in the least favored regions; however, the rates at the end are similar. In fact, 50% of the duration of the conflict is generated in year four for the most advanced region, while the least advanced region achieves this in year six. This is, the conflict diminished in the last years for both regions, but at different rates.

Next, following Llopis et al. (2022), we used a reduced random effects model with panel data, which shows the likelihood of conflict termination. We assume that the unobserved heterogeneity follows a Weibull distribution because we can estimate both the relative event rates and the relative extent of the event in survival time (Carroll, 2003). It is important to notice that in proportional hazard survival models, the baseline hazard $h0(t)$ can be attributed to the course of time, after controlling for covariates and unobserved heterogeneity. In other words, this is the number of continuous years of the conflict in the survival equation (Llopis et al., 2022). Specifically, the hazard function is as follows:

where $h(t,hom⁡,Xit−1)$ is risk of conflict continuation; $h0(t)$ is the baseline hazard approximated by conflict survival, and $Xit−1$ are the lagged covariates that include a selection of social and economic variables such as rural rate, number of political parties, Gini index, coal deposits, primary and secondary education coverage, unemployment rate and oil royalties. Finally, the lagged homicide variable was included to capture persistence and it is controlled for region (Gómez, Rivera-Lozada, & Uni, 2023) ^[6]. Table 2 displays the main results.

The outcome shows that the risk rate of conflict perpetuity is positive for all covariates. There is evidence in favor of a positive dependence duration pattern (the probability that the conflict period ends increases over the years). Despite this, the impact can be low or high. For regions with persistent homicides, high rates of rural population and medium educational coverage, the probability of ending the conflict increases very slowly over the years.

In the opportunity dimension, coca crops, the presence of mineral resources such as gold or coal, as well as the rural population and unemployment rates turn out to be statistically significant and have a direct relationship with the probability of occurrence of the event. That is, they all drive the probability of recurrence of the armed conflict in Colombia. However, in this group, oil royalties per capita are not significant, which indicates that it is unknown whether the territorial units that receive this payment have a higher probability of prolongation of the conflict. In general, it is evident that the opportunity dimension is key in the analysis because rural areas with lootable natural resources and little chance of employability are regions where the prolongation of the conflict is high.

In the political and institutional dimension, the effective number of political parties are significant and positive, which shows that a state with high political fragmentation (multiparty) has a higher probability of conflict recurrence compared to those that are not. On the other hand, as expected, the net coverage of secondary education shows that territorial units with better secondary education coverage have a lower probability of persistence of conflict. Conversely, those territories with lower educational levels (net coverage of primary education) show greater persistence. Thus, the political and institutional dimension also contribute to the explanation of this problem.

In the grievance dimension, poverty is significant and has an inverse relationship with the probability of occurrence of the event, which indicates that territorial units with an incidence of extreme monetary poverty have a lower risk of having an extended conflict over time. On the contrary, regions where GDP per capita and income inequality are high, the possibility of persistence of the conflict is higher. To sum up, as the literature shows, inequality, not poverty, is the variable that increases the probability of prolonged or recurrent armed conflict.

The survival model results should be taken with caution because we have few observations (263) and the KM survival function calculations reduce them significantly due to the use of time segments. Nonetheless, to check the model robustness a two-sided log-rank test is implemented (under Freedman approach) to detect the reduction in the risk index of the experimental group. This test computes sample size, power or effect size for survival analysis by comparing survivor functions in two groups (in our case, developed and less-developed states).

To achieve a power of 80% of this test (with alpha = 5%) and detect a 50% reduction in the hazard of the experimental group, the required number of events is 72 (failures). Because all states experience an event by the end of the study, the number of states required to be recruited to the study is equal to the number of events.

According our results, we achieve 54 failures (27 events and sample sizes with n = 8 per group), so the power of our test reach 70% (with alpha = 10%), detecting a 40% reduction in the hazard of the experimental group. All in all, we can affirm that our model is not so far form the ideal standard statistical numbers despite our few observations.

The survival model is a reduced model, in this sense; other important covariates belonging to the dimensions previously considered have yet to be included. For this reason, a parametric model is implemented in the next section to further deepen the analysis.

In light of Collier and Hoeffler (2004) modeling, to capture the likelihood of incidence of the armed conflict in Colombia, a discrete choice model (Probit) with panel data for 23 states in 2008–2018 was implemented ^[7]. In this sense, a balanced panel of 253 observations was obtained. In addition, to move away from Collier and Hoeffler's (2004) modeling, we address possible endogeneity via simultaneity by lagging the covariates by one period (Gomez et al., 2022) ^[8].

Furthermore, the model is estimated by random effects method since we consider that the individual effects are not related to the explanatory variables, and because there is no conditional fixed effects probit estimator for panel data. Although, there is an unconditional fixed effects probit estimator (that is, with dummy variables), the bias of the parameters increases when there are few periods, as in our case. On the other hand, there are problems to proper compute standard errors and the simulations require powerful computers. Greene (2004) corroborates these claims using Monte Carlo simulations.

The information used in the empirical model is by state and not by municipality due to several factors. Firstly, the armed conflict in Colombia does not take place in the capitals but, conversely, is intense in rural areas and low in large cities (Guzmán, Borda, & Luna, 2019; Fernández, 2012), especially where there is land ownership is concentrated (González & Carrillo, 2017). Secondly, this is reinforced because three quarters of the country's municipalities are predominantly rural (Parra-Peña, Ordóñez, & Acosta, 2013). Thirdly, the information for the main variables of this study is only available for the most important and/or large cities (23 according to the official statistics department, DANE) while for the rural municipalities, where violence prevails, information is very scarce or nonexistent (such as for the capitals or municipalities of Arauca, Casanare, Vichada, Guainía, Guaviare, among many others).

Lastly, the states in this country are not administrative units large enough to imply a high degree of heterogeneity within them. Indeed, rural areas in Colombia are characterized by being similar in aspects such as social and economic inequality (Cárdenas Pinzón & Vallejo Zamudio, 2016), lack of infrastructure (roads, public services, schools, hospitals, among others) or great natural wealth (precious metals, oil, rivers, among others). The model is as follows:

The dependent variable is homicides (hom). As we mentioned before, this is a dichotomous variable and therefore if the number of homicides in a state (i) exceeds the average number of homicides in the sample (23 states) in year (t), then it is equal to 1 and 0 otherwise. The vectors X, W, Z correspond to the opportunity, political and grievance dimensions, respectively. The time control variable (year) is a dichotomous variable that assumes the value of 1 for a particular year and 0 for the rest, and try to capture the possible fixed effects of macroeconomic shocks such as inflation, economic policies such as tax reforms or social policies as the peace agreement.

On the other hand, the region control variable is also dichotomous and assumes the value of 1 for a particular region and 0 for the rest and tries to capture the fixed effects of social and economic inequality of the territories, possible cultural diversities or the problem of land concentration. The idiosyncratic error $vit$ is composed of $(ai+εit),$ where the term $ai$ represents the unobservable individual effects and $εit$⁠, the stochastic errors, which are assumed to be well behaved. Finally, $Φ(∙)$ represents the Normal cumulative density function.

In the group of variables associated to opportunity dimension, there are coca crops (coca) because the NSAA find in the illegal production and trade of this natural resource an important source of financing and construction of illegal economies, therefore it constitutes an opportunity to impulse the conflict (Díaz Escobar & Sánchez Torres, 2004). In other words, there is a direct correlation between coca crops and armed conflict in Colombia (Vázquez, 2009), and therefore it is expected that the number of hectares of coca in the states is positively related to recidivism and prolongation of the armed conflict.

In the case of gold, the situation is similar, since to finance the war, guerrilla groups, FARC dissidents and paramilitaries control the extractivist chain of this metal through extortion and/or assassinations of social leaders and former guerrillas socially reinserted (Rettberg, Cárdenas, & Ortiz-Riomalo, 2017). Other additional factor is the government laxity in recent decades in obtaining mining titles (exploitation permits) from local and foreign nonstate agents (Ortiz-Riomalo & Rettberg, 2018), who affect the distribution of land by the commercialization of land and changes in land use of the agrarian reform. Taking into account all the above elements, we expect a direct relationship between violence and gold mining.

For the extraction of coal, the story is also similar to the two previous cases, since the economic interests existing in this mineral are very strong as it is the second export product in Colombia (Ángel, 2018). This generates competition over the sovereignty of the territories not only by groups outside the law but also by private entities. Moreover, the extractivist economic development model promoted by the local government explains the strong relationship existing between the deposits of coal and armed conflict (Sandoval, Marín, & Almanza, 2017). In this sense, we expecting violence will increase when the exploitation of coal is greater.

In the case of oil royalties (oil_royalties), as a natural resource, it affects the viability of rebellion in various conflicts around the world, and in Colombia, this natural resource represents an opportunity for financing and looting for NSAA. Ross (2006, 2015) and Collier et al. (2003) pointing out countries with oil reserves and dependence on or the presence of abundant natural resources (as in this case also with gold and coal) are more likely to break out and prolong civil war. According to González (2003), in the territories under the control of insurgent groups, local orders emerge that regulate economic activities, the participation of income and income generated, and influence the allocation of budgets and distribution of royalties. In this sense, Sánchez and Chacón (2005) show that the dispute over local power by the NSAA originates from access to public finances.

In other words, the states that receive royalties for the extraction of the oil resource in their territories are more likely to reoffend or prolong the conflict, since they are the focus of co-optation in public finances from the royalties of the resource (Rettberg & Prieto, 2018). In this order of ideas, the sign associated with this variable is expected to be positive, indicating that the states where there is extraction of oil resources (and others such as gold and coal) and with royalties associated have a greater probability of recurrence and perpetuation of the conflict as the mineral wealth and public finances in the state are a source of financing and extortion for the NSAA.

The rate of rurality (rural) is related to the sociodemographic configuration of the Colombian states and shows the number of the rural population over the total of each state living outside the capital cities. In Colombia, several factors explain the linkage between rurality and conflict. First, economic development models have historically institutionally privileged urban areas to the detriment of rural ones, which has motivated unforced displacement toward the capitals, becoming a source of conflict (Vergara et al., 2014). Secondly, the problem of land distribution in this country and the government's solution (known as Agrarian Reform, or more recently, Integrated Rural Development; Lleras Restrepo, 2012) which has shown ineffective and unsatisfactory results (Berman Arévalo, 2019; Ossa, 2014). Third, illicit crops, which are carried out in rural areas and are more profitable for farmers (Mendieta Daza, 2020), since the prices of legal crops are relatively lower due to the high level of intermediation, lack of producer association, volatility of demand and the climate uncertainties (Jiménez-Rodríguez, Gamboa-Suarez, & Vergel-Pérez, 2020). With all above, we considering states with higher rates of rurality provide a military advantage to the NSAA, and therefore, with higher rates of rurality, there is a greater probability of armed conflict take place.

On the other hand, the unemployment rate (unemployment) is an indicator of opportunities arising from an atypically low cost; that is to say, NSAA recruits must be paid and their cost may be related to income lost by enlisting as a rebel. Rebellions can occur when foregone income is unusually low. Another argument with the same effect is the lack of job opportunities (whether formal or informal) in rural areas (Otero, 2019), because the NSAA salaries/income may be more attractive for them since they fluctuate between 2 and 5 minimum salaries depending on the military range; wages unattainable in a legal or informal rural jobs/activities in Colombia. Therefore, the parameter associated with the unemployment variable is expected to be positive, as when there are precarious employment conditions (lost income is atypically low) the probabilities of prolonging the conflict are greater.

When considering the forced recruitment inherent to the harshness and cruelty of the conflict in Colombia, the net education coverage variable is a proxy variable for the institutional dimension, since the states with better coverage in secondary education have a lower probability of risk of prolongation of the conflict.

For the representative variables of the political and institutional dimension, this document shows the importance of the actors, media and channels that contribute to the consolidation and strengthening of the NSAA. Following Laakso and Taagepera (1979), a representative category of analysis is the index of the effective number of political parties (pol_part), which measures fragmentation and multiparty electoral competition. This variable results from the number of political options weighted by their electoral weight in terms of their vote; taking 1 as the minimum value and can be extended, theoretically, up to the total number of actors in electoral competition. Variations in the pol_part indicate changes in the distribution of votes between parties; thus, an increase means a greater number of electoral offers, which tends to blur the identity and ideology of parties, turning them into similar alternatives with slightly different political programs (Crisp, Olivella, & Potter, 2015, p. 48).

The pol_part index results are classified according to the typology of Hoyos (2005). That is, strong one-party if $1≤pol_part<1.6$⁠; attenuated one-party party if $1.6≤pol_part<2.0$⁠; two-party system if $2.0≤pol_part<2.7$⁠; and multiparty if $pol_part≥2.7$⁠. Hence, according Hoyos (2005), only one state presented a single-party system, and the rest were divided into states with two-party and multi-party systems. In this sense, pol_part is a dichotomous variable that takes the value of 1 if the state is multiparty and 0 otherwise. A high value of pol_part captures the existence of a greater number of electoral offers (a change in the distribution of votes between parties). In this sense, it should be considered that the political reform of 2003 implied the deinstitutionalization of the party system at the regional level, making relations easier and less costly in political and judicial terms. Thus, it is expected that the sign associated with the parameter of the dichotomous variable pol_part is positive, showing that a multiparty state has a higher probability of recurrence of conflict compared to those that do not. This is so, as highly fragmented states are a sign of possible articulations of conflict legal and illegal political actors, which are alliances that contribute to the territorial and financial strengthening of nonstate armed actors.

Another determining political and institutional dimension is the net coverage of primary (prim_cov) and secondary (med_cov) education. Collier and Hoeffler (2004) take these variables as indicators of foregone income; however, they show that the proxy is open to other interpretations, since “education can affect the risk of conflict through changing attitudes”. In addition, the low or null coverage in education implies in the short and medium term less possibilities of accumulation of human capital and therefore of socioeconomic progress (Becker, 1993), being the enlistment in the NSAA an alternative to procure an income. On the other hand, without education due to lack coverage, the possibilities of migrating to capitals cities or large ones are more difficult and therefore young people are exposed in rural areas. Both paths leave infantrymen vulnerable to forced or voluntary recruitment. According to National Centre for Historical Memory (CNMH) (2021) in Colombia at least 17,866 children and adolescents have been victims of recruitment and use in the armed conflict between 1958 and 2020. In this sense, following Mendoza Tovar (2021), as forced recruitment is part of the harshness and cruelty of the conflict in Colombia, it is considered in this research as representative of the institutional dimension. Thus, the states with coverage in elementary education have a high possibility of perpetuating the conflict, while those with greater coverage in secondary education have less probability of prolongation.

Initially, the representative variables in the grievance dimension include the incidence of extreme monetary poverty (poverty) as a representative of complaint or grievance. The traditional literature finds that poverty explains the conflict due to the inability of the State to satisfy the basic needs of many citizens such as education, health, security, employment, among others (Garrido, 2001). This increases the possibility that individuals become criminals, question the institutions and therefore form part of groups outside the law (Galindo, Restrepo, & Sánchez, 2009). However, recent literature finds that this hypothesis loses relevance to explain an increased probability of recidivism and perpetuation of the Colombian conflict, because the NSAA, with the aim of finding sources of financing, move to territories with higher levels of wealth, economically viable and highly economically active. Consequently, tempered by modern literature, we consider that territories with extreme poverty cease to be the epicenter of looting or extortion; hence, the sign associated with the parameter of the poverty variable is expected to be negative, indicating that poorer territories they are less likely to suffer the intensity of violence resulting from the prolongation of the conflict.

Regarding GDP per capita (gdp_pc), this shows the possible disagreements derived from the behavior of the economy. In this sense, an inverse relationship with the strengthening of the conflict is expected, since if in the previous periods the GDP per capita have grown the complaints about the development of the economy decrease and the risk of conflict. Indeed, many of the serious protests in 2021 for this country are due to the economy performance (growth, income distribution and development) in the recent past (Díaz Guevara, 2021). However, it is worth clarifying that this factor could also be part of the opportunity dimension, since as it is related to aspects of the economy such as unemployment, income or consumption, among others, it can be used by NSAA to generate and/or promote social and political instability. Under this premise, we expect an also negative sign. It is worth mentioning that the mathematical results do not change if this variable is included in one dimension or another, since they are all part of the explanatory variables of the stochastic model.

Lastly, income inequality is captured with the Gini coefficient (gini). Following Botello (2014), the lack of opportunities due to income gaps prevents social mobility, strengthening general discontent and therefore crime emerges as an alternative to climb the social pyramid. Other aspects that contribute to income inequality, such as the concentration of land and/or capital, state abandonment, labor informality or educational coverage, strengthen forced displacement, which generate or strengthen poverty traps that in turn drive the armed conflict (Sánchez-Torres, 2017; Rodriguez & Torres, 2009). This author considers that the regions in Colombia present different levels of social inequality; therefore, this also partially justifies the control of the stochastic model with dichotomous region variables.

The lack of information for some territories in Colombia does not generate bias in this model. This is so as the states with scarce or none information and therefore left out of this study (Amazonas, Arauca, Casanare, Guainía, Guaviare, Putumayo, San Andrés and Providencia, Vaupés and Vichada) show very similar socioeconomic characteristics not only among them but also with those of the less-developed region. In other words, the less-developed states represent well the absent departments in this research; consequently, there is no selection bias. In fact, the least developed and absent states are included in Colombian economic periphery (Ramírez & de Aguas, 2017).

To confirm this, a mean difference test is implemented on the variables HDI, homicides and income (GDP per capita) in 2012; the only year in which there is information for all states. The results are displayed in Table 3.

The results reveal that in none of the indicators considered do the sample means show statistically significant differences between the absent states and the less-developed ones, since the t-statistic is low, so we reject the null. This confirms our hypothesis that the less developed and absent states are part of the same group since they do not have different dependent characteristics. It is worth mentioning that these results are in line with those found by CEPAL (Ramírez & de Aguas, 2017).

Table 4 shows three specifications of the model (Equation (2)). The first column contains no control variables, the second column contains only the year, and the third column contains year and region. In general, the results show a better adjustment for the latter case, showing the relevance of controlling for external shocks and for the territory where the state is geographically located.

In the opportunity dimension, coca crops, the presence of mineral resources such as gold or coal, as well as the rural population and unemployment rates turn out to be statistically significant and have a direct relationship with the probability of occurrence of the event. That is, they all drive the probability of recurrence of the armed conflict in Colombia. However, in this group, oil royalties per capita are not statistically significant (and its impacts are strongly low, near to zero because of units of measure), which indicates that it is unknown whether the territorial units that receive this payment have a higher probability of prolongation of the conflict. In general, it is evident that the opportunity dimension is key in the analysis because rural areas with lootable natural resources and little chance of employability are regions where the prolongation of the conflict is high.

In the political and institutional dimension, the effective number of political parties is significant and positive, which shows that a state with high political fragmentation (multiparty) has a higher probability of conflict recurrence compared to those that are not. On the other hand, as expected, the net coverage of secondary education shows that territorial units with better secondary education coverage have a lower probability of persistence of conflict. Conversely, those territories with lower educational levels (net coverage of primary education) show greater persistence. Thus, the political and institutional dimension also contribute to the explanation of this problem.

In the grievance dimension, poverty is significant and has an inverse relationship with the probability of occurrence of the event, which indicates that territorial units with an incidence of extreme monetary poverty [percentage of the population with a household per capita income below the extreme poverty line by geographic domain (DANE, 2019, p. 15)] have a lower risk of having an extended conflict over time. On the contrary, regions where GDP per capita and income inequality are high, the possibility of persistence of the conflict is higher. Moreover, as the literature shows, inequality, but not poverty, is the variable that increases the likelihood of prolonged or recurrent armed conflict.

The Colombian armed conflict is strongly explained by the thesis on greed in regard to natural resources, which is directly related to the high homicide rates and continuous forced displacement throughout the Colombian territory and, hence, indicators of the persistence of the armed conflict. Consequently, there is abandonment of land, crops, uprooting and identity as well as economic dispossession for hundreds of thousands of Colombians.

There is a positive correlation between the greater concentration of property and the forced displacement of peasant communities in states with the expansion of agro-industrial crops such as African palm and large-scale mining. Thus, large agricultural and mining projects are located in the main expelling municipalities, such is the case of southern Guajira, Cesar, Urabá, Córdoba, Catatumbo/Norte de Santander, southern Bolívar, Buenaventura/Valle, Tumaco/Nariño, Putumayo, Caquetá, Guaviare, foothills of Meta, Arauca (Fajardo, 2015, p. 21), all states with persistent armed conflict during the period 2008–2018. In this sense, Figure 2 shows the panorama of forced displacement in Colombia.

Forced displacement has been a strategy historically used in the Colombian territory for the dispossession of land from hundreds of thousands of peasant families, configured in an agrarian reform as it means the change of properties and land titles that derives in the concentration of property for commercial agriculture. This Colombian phenomenon of accumulation by dispossession is a land grabbing modality that responds to economic and political interests that pressure the eviction of the population in the territories (Vega, 2012).

The Centro Nacional de Memoria Histórica (CNMH) (2013) stated that 2,729,153 people were displaced during 2003–2012, as a result of the territorial expansion strategy of paramilitary groups, deficiencies in the demobilization processes enshrined in Law 975 in 2005, the regrouping and rearmament of some demobilized people. Also, the military offensive of territorial recovery by the state in compliance with the Democratic Security Policy; and the massive and indiscriminate planting of anti-personnel mines by the FARC. Forced displacement and land dispossession caused the appropriation of uninhabited lands through legal resources to formalize the taking of land or the purchase at low cost, taking advantage of the vulnerability of the market and, in other cases, resorting to violent mechanisms that lead to dispossession (Centro Nacional de Memoria Histórica (CNMH), 2013).

In Colombia, accumulation by dispossession liberates a set of assets, including labor force, at very low or even zero costs (Harvey, 2005), while dispossession links economic activities, land appropriation, assassinations, massacres, torture and forced displacement as a vehicle for land concentration by businessmen who promote the accumulation of capital in the countryside (Vega, 2012). Thus, it is clear that homicide and forced displacement due to the accumulation by dispossession of land resources aggravates the structural conditions in which the armed conflict develops. Due to the transformations in land use, there is a long-term agrarian change: overlapping of monocultures, extensive cattle ranching, mining and oil exploitation, and agro-export, this demonstrates that these phenomena make up a process of accumulation by dispossession that takes advantage of the armed conflict to mask these realities (Vargas, 2021).

The permanent Colombian agrarian conflict of the last sixty years is the continuity of the struggles of settlers, indigenous, peasant and Afro communities from the beginning of the nineteenth century in search of justice and social equality. The systematic expropriation of peasants' lands over time aims at accumulation by dispossession, as well as expanding the frontiers of agricultural production to plant crops and monocultures for export, agro fuels, increasing grazing land for livestock, extraction of mineral wealth, construction of dams that imply the monopoly of water (Vega, 2012).

Now, the nonstate armed actors and the state are co-responsible for the whole process of accumulation by dispossession and its disastrous consequences: massacres, forced displacement, concentration of land ownership. It contributes to the legalization of land and nature expropriation, through laws or statutes that help with prescription times; solving the cases of falsa tradición (it occurs when a right or real estate is transferred to anyone without being the owner of the property). Besides, the validation of nonoriginal titles as all these mechanisms solve disputes in favor of large landowners and those who have appropriated the land in the last ninety years (Vega, 2012). In the Colombian case, the State created institutions, conditions and alliances for rural dispossession and the allocation of property rights in certain hands because of coalitions to obtain territorial control and promote the process of accumulation by dispossession (Vargas, 2021; Bedoya & Rocío, 2019).

The approaches presented in the framework of this research evidence that in Colombia land is a covetable resource and has an impact on the perpetual violent actions. Thus, the Greed-Grievance model of Collier and Hoeffler (2004) addresses land as a source of grievance and not as a source of financing, looting or extortion. The model incorporates the measurement of income inequality with the Gini coefficient and asset inequality with the Gini coefficient of land ownership, in terms of grievance and not of greed because they do not have statistical significance and, hence, they do not explain the risk of outbreak of civil war. However, in Colombia, access to land continues to be a cause of both grievance and greed, which is currently reflected in the social protests of peasants, trade unionists, human rights defenders and social leaders who demand justice and equity for access to this resource.

In view of the above, an effective policy to end the conflict must reduce the economic incentives for war and prioritize a rigorous control of production, resource extraction and derived royalties; that is, it must reduce and control the sources of financing of the NSAA. By virtue of the above, one of the points of the Peace Agreement (2016) most attacked by opponents of the Agreement has been the comprehensive rural reform that makes access to land more equitable, decreases income inequality and unemployment and improves access to education. Additionally, it restricts the sources of looting and extortion that finance the activities of nonstate armed actors and political actors interested in the prolongation of social disorder. However, mitigating violence that stems from the fight for these resources, between the NSAA, businesspersons, multinational firms and legal and illegal rural elites, implies looking at the land as a resource. Thus, we can mention that it contains natural resources that are financing opportunities for the NSAA (gold, coal, oil royalties and coca cultivation, among others). They are object of productive and symbolic covetousness for political actors and rural elites that have illegal articulations with armed actors, with the purpose of dispossessing inhabitants of their land to exploit and loot natural resources. No less relevant is the impact of international drug policy, which can alter the structural conditions of the market in terms of incentives, costs and legal repercussions, all factors that affect the behavior of armed actors.

This research investigated the factors that determine the perpetuation of the armed conflict in Colombia based on the following dimensions: opportunity, political and institutional dimension, and grievance. For this purpose, we implemented a quantitative methodology based on survival functions, risk and discrete choice models (Probit). Finally, it offers an interpretation of what has historically been the armed conflict in Colombia and what is required by the state to cut off the sources of financing of armed groups and thus, one of the determinants of the continuity of the conflict.

The research offers empirical evidence that identifies the opportunity dimension as a strategic one to understand the perpetuation of the armed conflict in Colombia. Specifically, it proves the thesis on greed based on the existence of natural resources that can be looted in rural areas with few employment opportunities, which enhances the prolongation of the conflict. Therefore, the greed thesis, in addition to being a source of aggravation for the prolongation of the Colombian armed conflict, finances the rebellion to ensure land grabbing and, thus, monopolize the ownership of lootable natural resources.

The study divided the country into two large regions depending on the level of development and economic growth. This makes it possible to affirm that the behavior of the variables analyzed at the national level generally coincides with the pattern shown by the same variables in the least favored regions. Furthermore, in both regions, there are characteristics that could perpetuate the conflict, such as natural resources (gold, coal or oil) in the least favored regions and high levels of wealth in the most prosperous regions, both conditions attract violent groups in pursuit of economic benefits.

The perpetuation of the Colombian armed conflict is strongly explained by components of the political and institutional dimension. In particular, the research shows that states with high political fragmentation (multiparty) are more likely to continue with conflicts than those that are not. These results highlight the regional tensions resulting from the operación avispa, by which Colombian political parties gain access to several seats in collegiate bodies by taking advantage of the partitioning of lists within the electoral quotient and residue system, exposing territories to intense struggles for control and access to territorial resources (Leiteritz, Nasi, & Rettberg, 2009; Peña, 2014).

The used model shows that the feasibility hypothesis has explanatory power in the continuity of the Colombian armed conflict. The evidence indicates that territorial units with extreme monetary poverty have a lower risk of an extended conflict over time. Meanwhile, regions with higher GDP per capita, together with income inequality indicators, have a greater propensity to perpetuate the conflict. These results are corroborated by the Kaplan–Meier survival functions and the proportional hazard model, which reaffirm that social and economic inequality are the sources that feed the conditions for prolonging rebellion because they contribute to the opportunity structure that makes the conflict militarily and economically viable in Colombia.

Finally, the lack of information for each state in Colombia prevents a much more holistic analysis of this problem, as in the case, states of the Orinoco and Amazon regions where the armed conflict is recalcitrant are excluded due to this cause.

The authors wish to thank the Universidad del Cauca for the support in the preparation of this article.

The supplementary material for this article can be found online.