Portal de Periódicos da CAPES

The persistent Colombian conflict: subnational analysis of the duration of violence

2012; Taylor & Francis; Volume: 23; Issue: 2 Linguagem: Inglês

10.1080/10242694.2011.597234

1476-8267

Juan F. Vargas,

Terrorism, Counterterrorism, and Political Violence

Abstract Focusing on the Colombian armed conflict, this paper develops for the first time a within-country analysis of violence duration. It examines a wide range of factors potentially associated with violence duration at the municipal level for the period 1988–2004, including geographic factors, economic and social variables, institutional characteristics, victimization variables and government intervention. It individuates the most robust correlates of the persistence of localized conflict, both across specifications and using different econometric models of duration analysis. Results suggest that violence in Colombia is more persistent in places where illegal rents are available. Better quality institutions and a more active military are in turn associated with shorter conflict episodes. Keywords: ColombiaConflict durationSurvival models Acknowledgements I am grateful to Catherine Rodriguez, Jurgen Brauer, participants of the workshop on the Colombian conflict held at Universidad del Rosario, Bogota, Colombia, on 16th November 2010, and one anonymous referee for useful comments and discussion. Notes 1For a detailed account of the conflict see Rabassa and Chalk (Citation2001). 2There is a large regional and temporal variation in patterns of violence in Colombia. These take the form of both uncontested attacks, generally carried out by illegal groups (like incursions, bombings, road checkpoints and massacres), and clashes against government forces. The bulk of clashes or fire exchanges in the Colombian conflict involve an illegal group (generally a guerrilla organization but in some cases right-wing paramilitaries) against government forces (military or police). However, there are some episodes of clashes between guerrillas and paramilitaries, or clashes between different organizations within the guerrilla or within the paramilitary. 3It is worth noting that in Colombia violence does not present a definite seasonal pattern within a given year. This does happen in other conflicts (like Afghanistan) due to the severity of inter-season climate changes. 4A similar strategy is often used in the cross-country literature on the causes of war. When violence recurs shortly after the conflict ended (say, within five years –Toft, 2010), the previous war is said to have recurred and no new war is coded. 5CERAC maintains a unique event-based dataset that covers over 21,000 conflict-related incidents over the period 1988–2004. For each event, the dataset records the date, location, type, perpetrator, and victims involved in the incident. In terms of type, it records whether the incident was an uncontested attack, carried out by an identified armed group against a specific military or civilian target, or a clash, which involves an exchange of fire between two or more groups. In terms of perpetrators, it records whether attacks were carried out by the guerrillas, the paramilitaries or the government, and details the groups involved in a clash. In terms of victims, it reports the number of casualties separately for combatants and civilians (see Restrepo, Spagat, & Vargas, Citation2004 for a complete description of the CERAC dataset). 6Violence was ongoing in 230 municipalities in December 2004, the last period covered by the conflict data used in this paper. This constitutes 9% of the total violence episodes analyzed for the period 1988–2004. 7Of course, it would be wrong to assume either that violence ended in 2004 altogether or that it will never end in these municipalities. 8In practice the log of the likelihood is easier to maximize and gives the same estimates for β and its unceritainty. I estimate the log-likelihood associated with (3), which is: . 9Although restrictive, there are ways out of this assumption. Collier et al. (2004), for instance, use a 'piecewise' hazard approach that separates out the hazard function into a 'baseline' hazard that does not depend on covariates and does not vary by country but is only a function of time, and a covariate-varying component: h(ti ) = hB (t) exp[−μi (Xit β)]. The baseline hazard consists of dummy variables that capture year-specific deviations from the constant hazard. 10Both the Weibull (and hence the exponential) and the lognormal distributions are special cases of a distribution called generalized gamma, so ideally one would like to fit the latter distribution hence making a less subjective and more data-based choice of the right duration model, an approach followed by Box-Steffensmeier and Jones (Citation2004) for the case of cabinet duration. However, with two ancillary parameters, the generalized gamma is computationally unstable and I could not consistently estimate it for most of the specifications reported in the paper. 11Since there is no distribution for the length of violence, Ti , underlying the Cox approach, expected duration cannot be calculated in this case. 12For dichotomous variables the marginal effect is computed from 0-to-1 changes of the variable. 13Recalling that the aim of this paper is to present a thorough review of the local-level correlates of violence length within a country experiencing a civil war, no attempts to estimate causal effects is made. With this caveat in mind the estimated cardinal magnitudes should also be interpreted with caution. 14In the case of this particular regression model which includes dummies of coca growing and the availability of mining royalties the baseline duration is the average violence duration in municipalities that neither grow coca nor receive royalties. 15Colombia is the world's number one producer of coca (UNODC, Citation2010), used to manufacture cocaine, as well as one of the largest producers of poppy plants, used in the production of heroin. 16The next subsection looks more closely at the robustness of the single group correlates. 17With financial and military aid from the US, from year 2000 an active military campaign has been promoted by the Colombian government to fight insurgents. This aid package is known as Plan Colombia and its aim is to cut down the supply of drugs dispatched to the US. 18The estimates of σ in all the models presented in Table 2A suggests that it is not statistically different from 1 (its log is not different from zero) which implies that fitting an exponential distribution instead of the Weibull would have mischaracterized the non-flat shape of the baseline hazard. 19Because the goal of Table 3 is to look at which of the significant variables of the previous exercise survive the inclusion of different set of controls, I do not report the hazard-rates marginal effects. I will come back to these in the final analysis of the next section. 20The over 1,000 municipalities are grouped in 33 departments which in turn are grouped in five regions. 21Recall that this is sensitive to the arbitrary 12-month threshold used to code recurrence of violence (see section 3.1). 22This is in line with the emphasis of the 2011 World Bank's World Development Report which suggests focusing on institutions to help prevent violence.