Child Soldiering in Colombia: Does Poverty Matter?

Abstract

Child soldiering remains a challenge for the international community, and non-state armed groups are the most persistent offenders, but its drivers are poorly understood. Recent contributions suggest that decisions by armed groups are the key to explain child soldiering and that contextual variables are less relevant. This article exploits the availability of subnational, longitudinal data on child soldiering in Colombia, where insurgents and private militias have recruited children at least since the 1990s. The analysis shows that child recruitment is more likely in poorer municipalities, with limited access to education, and where coca crops are grown.

Notes

1. In this article, the term ‘armed group’ refers to non-state armed groups.

2. Paradoxically, in military circles children are perceived as anything but ingenuous. See Kilcullen (2010, p. 40) or United States Army (2007, p. 296).

3. For instance, research shows that assumptions on children’s naivety or moral immaturity may be true for children under six years of age but not for teenagers (Berk 2003, Heyman and Legare 2005, Hoffman 2000, Sperber et al. 2010).

4. In doing this, they ignore that in terms of cognitive skills teenagers (i.e., above thirteen) are way more similar to adults than to younger children (Moshman 2011).

5. Werner Herzog’s documentary “The Ballad of the Little Soldier” (1984) also provides vivid evidence on child soldiering during the Nicaraguan Revolution.

6. Colombia ratified the Convention on the Rights of the Child in 1991, the Additional Protocol II to the Geneva Conventions in 1995, the Rome Statute in 2002, and the ILO Convention No. 182 of 1999 and the Optional Protocol to the Convention of the Rights of the Child on the involvement of children in armed conflict in 2005.

7. In Colombia, the reintegration of demobilised ex combatants is assigned by law to different agencies depending on the age of demobilisation: adults go to the Colombian Reintegration Agency (ACR), children to the Family Welfare Institute. The judiciary deals with guerrillas captured in action. To our knowledge, the study by Sepulveda et al. (2013) was the first to merge the three databases. Unfortunately they did not release the dataset.

8. One notable exception is the study by Llorente et al. (2005).

9. We are thankful to Ana Arjona for disclosing these figures for us in a personal communication.

10. This is consistent with the recent finding that, in Colombia, the ‘the most powerful predictor’ of the armed group someone will join is the ‘political leaning’ of their family. See Ugarriza and Craig (2013).

11. In their analysis, Sepulveda et al. (2013) only included variables correlated with child recruitment but, apparently, poverty was not one of them. Restrepo (2013) included poverty, which turned out to be significant, but the choice of an OLS model raises doubts about his conclusion.

12. The proportion of demobilised combatants who were recruited as child soldiers is in the range of 44–50% for former FARC members and in the range of 35–40% for former paramilitaries. These estimates are based on Springer (2012) and on figures obtained by the authors from the Colombian Reintegration Agency. For descriptions of the organisational features of these groups see Ferro and Uribe (2002), Duncan (2007), and Gutiérrez (2004, 2008).

13. Displaced households are far more likely to live in poverty than the average Colombian household (63 vs. 25%, respectively); approximately one in three displaced households live in extreme poverty. See DANE (2015).

14. Annual statistics are available at http://www.biesimci.org/Documentos/Documentos.html, accessed on July 30, 2015.

15. Although this could be the effect of omissions in the data, we believe it reflects the clandestine nature of irregular organisations.

16. The Noche & Niebla series provides detailed narratives of events related to armed conflict and political violence in Colombia since 1996. It is available at http://www.nocheyniebla.org Access to Cerac’s data can be requested at http://www.cerac.org.co.

17. Zero-inflated negative binomial panel models are unusual in the literature, and unavailable in Stata 13, so we opted for a hurdle model, which also allows for zeros and positive counts to come from different data-generating processes. See Cameron and Trivedi (1998).

18. Using a fixed-effects model in a case like this would amount to ‘throwing out the baby with the bath water as too many observations with relevant information would be dropped. See Beck and Katz (2001).

19. The binary process accounts for excess zeros, that is, for the absence (not the occurrence) of the phenomenon. See MacDonald and Lattimore (2010).