Paramilitary Violence in Colombia: A Multilevel Negative Binomial Analysis

ABSTRACT

Although Colombia is well known for its persistent leftist guerrilla conflict, the country also suffers from paramilitary violence. This study examines the potential factors related to persistent paramilitary violence in the form of human rights violations. How has paramilitary activity, and its causes, changed over time? Why does it persist in some areas after Uribe’s demobilization process but not in others? We use multilevel modeling to explore the determinants of paramilitary human rights violations. A varied range of aspects potentially associated with the paramilitary presence at the municipal level for the period 2002–2015, such as state presence, resources, greed, grievances and conflict are analyzed. The study uses information about paramilitary human rights violations from the Centro de Investigación y Educación Popular (CINEP). Results suggest that the demobilization process reduced the initial paramilitary motivation to fight against leftist guerrilla. However, other factors such as coca cultivation or ranching remained significantly related to the paramilitary activity. The analysis at the municipal level provides clear warnings for continued violence cycles threatening any undergoing or future peace processes or demobilizations and calls for a more nuanced concept of state capacity to understand paramilitary violence.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. For a micro level analysis, see Arjona A.M., Kalyvas S.N. (2012) Recruitment into Armed Groups in Colombia: A Survey of Demobilized Fighters. In: Guichaoua Y. (eds) Understanding Collective Political Violence. Conflict, Inequality and Ethnicity. Palgrave Macmillan, London.

2. Las Cooperativas de Vigilancia y Seguridad Privada (Convivir) were created during the government of Cesar Gaviria through the Decree Law 356 of 1994 and, later, put into operation during the government of Ernesto Samper.

3. Structural break analysis on the time series showed two main breakpoints in the year 2004 and 2010 (Figure 1). These breaks are consistent with what was found by the Centro Nacional de Memoria Histórica. CNMH (2016, 196) which established that ‘for the first stage after the demobilization (2006–2010), the forced displacements made by the GAPD decreased in intensity and geographic expansion.’ An optimal breakpoint is calculated under Bai & Perron (2003) algorithm and represents a shift in the time series trend that explains a potential change in the data generation processes.

4. CINEP categorizes the human rights violations based on the International Humanitarian Law. Specifically those mentioned in the Geneva Convention of 1949 and Protocol I of 1997.

5. CINEP is affiliated with the major international groups such as the Consejo Latinoamericano de Ciencias Sociales (CLACSO), Asociación Latinoamericana de Organizaciones de Promoción al Desarrollo A.C. (ALOP), International Land Coalition (ILC), International Coalition for the Responsability to Protect (ICRtoP), Red de Centros Sociales de la Compañía de Jesús en América Latina (CPAL), Plataforma DHESCA, and the Global Ignatian Advocacy Network (GIAN). Nationally, they are affliated with the Corporación de Desarrollo y Paz de Magdalena Medio (CDPMM), Corporación de Desarrollo y Paz del Oriente Antioqueño (PRODEPAZ), Corporación de Desarrollo y Paz del Piedemonte Llanero (CORDEPAZ), Corporación Viva La Ciudadanía, and Red Nacional de Programas Regionales de Desarrollo y Paz (http://www.cinep.org.co/Home2/institucion/nosotros.html).

6. Gutiérrez Sanín and Wood (2017) emphasize the importance of patterns of violence. Accordingly, from our CINEP data, we identified the most common forms human rights violations in this period: killings, threats to groups of individuals, threats to individuals, collective displacement, and intentional injuries to civilians. Nonetheless, we removed three categories related to attacks on a military target or perfidy and six categories related to individual or collective threats.

7. Author translation ‘ …, la presencia paramilitar en Puerto Asís, Orito y la Hormiga coincidía con un mayor cultivo de uso ilícito de coca, así que después de la desmovilización del Bloque Sur Putumayo se mantuvo activa una parte de esta estructura, ligada a la economía del narcotráfico en la región.’

8. Carlos Castaño et al. Letter to US Ambassador Anne Patterson 26 October 2001. (2001) ‘Nuestra Organización, las Autodefensas Unidas de Colombia, nació de la unión de disímiles grupos con intereses igualmente diversos, pero siempre orientados a la legítima defensa de la vida y los bienes de los ciudadanos ante la evidente insuficiencia y falta de voluntad política del Estado colombiano para combatir a los grupos de guerrillas marxistas’ Author translation. See also Carlos Castaño, ‘Quien sabe para dónde vamos’ (12 November 2001 Dipom 2001) ‘Recuerdo que hace muchos años, cuando iniciaba mi lucha antisubversiva, solo pensaba en ejecutar guerrilleros, y a decir verdad, poco o nada me preocupaba de otra cosa. No tenía nada mas en que pensar, ese era mi mundo, desconocía el otro, mi ignorancia era absoluta.’ Author translation. Accessed at www.colombialibre.org in 2006. Accessed at www.colombialibre.org in 2006.

9. New departments (nuevos departamentos) created in 1991 do not report unemployment and GINI coefficients. These departments are sparsely populated and they represent about 5% of the total population (Dirección de Metodología y Producción Estadística – DIMPE 2013, 2). In addition, GINI coefficients for 2006 and 2007 were not reported by DANE because of problems with the survey. We used basic linear interpolation to impute values for those missing years.

10. Notwithstanding, a confirmatory analysis of non-normality used OLS panel fixed effects model to check if OLS assumptions do not fit with the data nature. Panel estimates and residual tests to check for serial correlation and cross-sectional dependence showed in Table A4 of the Appendix confirmed suspected OLS violations.

11. To support the direction of the relationship between leftist guerrilla and paramilitary violence, we used a Granger test on the time series of the number of violent events of these two antagonist groups. Results of Granger causality for the conflict variable are shown in Table A1 of the Appendix.

12. The magnitudes of the coefficients that come from a GLMM negative binomial can be interpreted as Incidence Rate Ratios (IRR) if they are exponentiated. We decided to report the logarithmic results to keep the interpretation of results easy to understand. Given that our dependent variable, the Human Rights Violations, is positive and discrete, we prefer to leave the practical significance in terms of ‘negative relationship’ (when IRR < 1) or ‘positive relationship’ (when IRR > 1) to interpret factors that may produce an additional human rights violations compared to factors that might not produce an extra event.

13. See Table A7 in the Appendix.

14. We run a GLMM model that includes the spatial lag of human right violations. The results are shown in Table A6 in the Appendix. As it is shown the spatial lag is not significant, meaning that the model accounted for the spatial autocorrelation and confirmed with the residual tests.

15. Previous diagnostics on each category revealed singular dynamics and low levels of correlation. Each variable within categories has a unique contribution to the paramilitary activity. Correlation tests of significance are provided in Table A2 of the Appendix.

Additional information

Funding

This work was supported by the National Science Foundation [1541199].