In this paper, the concept of poverty is grounded in the capability approach (CA) framework pioneered by Amartya Sen. CA is a normative theory that evaluates the quality of life of individuals according to the choices available to them. Sen (1985, 1996) suggests that an individual’s quality of life can be evaluated by analysing the capabilities they have at hand to achieve valuable functionings such as state of health or education, participation in political deliberations, avoiding avoidable morbidity, or being happy. This evaluation occurs within the “capability space”, in which “capability represents various combinations of functionings (beings and doings) that the person can achieve” (Sen, 1992, p.40). However, when a capability set is restricted by externalities (such as armed conflicts), individuals face a smaller range of choices, leading to a restriction in the range of achievable functionings, and ultimately reducing their chances of achieving a dignified life (Nussbaum, 2011).

Seen through this lens, poverty is a problem of capability-deprivation (Sen, 1999). Furthermore, as the capability set indirectly reflects the freedom of choice a particular person enjoys, poverty can also be interpreted as the result of a lack of the freedom required to achieve valuable functionings (Sen, 1988). Since these functionings represent different states of being and doing, it is clear that the capability approach calls for a multidimensional measurement of poverty centred on the human being.

Antioquia is a Department (or province) located in the central northwestern region of Colombia. In terms of its contribution to GDP (around 13% in 2012, according to official statistics) it is second in importance only to Bogotá. Recent estimates by Colombia’s National Statistical Office, the Departamento Administrativo Nacional de Estadística (DANE), suggest that the population of the Department is close to 6,500,000, distributed across 125 municipalities; more than half of its population lives in nine municipalities in the Metropolitan Area of Medellín, in the Valle de Aburrá. For administrative purposes Antioquia’s municipalities are grouped into nine regions Readers may find it useful to refer to the map of the regional subdivision of Antioquia that is provided in the Appendix, as it is referred to in the Results section.

This study uses data on armed actions perpetrated by both illegal and legal groups in Antioquia between 1996 and 2010, and on forced internal migration (displacement) of persons expelled from their lands during the same period. The data was gathered by the Vice Presidency’s Observatorio de Derechos Humanos (Human Rights Observatory) and by the Noche y Niebla database maintained by the Jesuit NGO CINEP. Poverty was measured at household level for each municipality of Antioquia using the multidimensional poverty index (MPI), with the information being drawn from the government’s SISBEN database (cross-section 2012), which is used by several Colombian institutions as an instrument to target social assistance programs. Subsequently we provide a poverty measure for each municipality in Antioquia that is closer to Sen's capability approach.

The MPI, first proposed by Alkire and Foster (2011), provides a measure based on the capability approach. In the parlance of the MPI, each functioning is called a dimension. Interpreted as a direct method of poverty measurement,[3] which “shows whether people satisfy a set of specified basic needs, rights, or –in line with Sen’s capability approach– functionings (Alkire and Santos, 2013, p. 5)”, the MPI is not entirely new. Direct methods of poverty measurement have been widely implemented in Latin America by governments seeking to measure poverty through unsatisfied basic needs (UBN) indexes (Feres and Mancero, 2001).

Direct methods of poverty measurement rely on the dual cutoff method to identify the poor. First, direct methods determine the deprivations suffered by the population, namely the basic needs or rights that remain unsatisfied. Second, the poor are identified among the group of individuals who display any form of deprivation. Traditional direct poverty measurements generally rely either on the union or on the intersection identification methods. The union method regards as poor individuals with at least one deprivation or unsatisfied need, while the intersection approach categorises as poor individuals who experience deprivations across all dimensions. Thus, a key contribution of Alkire and Foster's MPI (2011) resides in the development of a new documentation procedure that “identifies the poor by counting the dimensions in which a person is deprived” (Alkire and Foster, 2009, p. 1). Moreover, the MPI satisfies the property of dimensional monotonicity, which states that “if a poor person becomes newly deprived in an additional dimension, then overall poverty should increase” (Alkire and Foster, 2009, p. 12).

Following the method developed by the 2014 UNDP Human Development Report, the approach to MPI used in this paper is comprised of three dimensions: health, education, and standard of living, which represent socially accepted valuable functionings and are proxied by nine indicators. These indicators do not necessarily reflect specific functionings, but do at least reflect circumstances conducive to the realisation of a given UNDP functioning. Moreover, data limitations largely explain why it was only possible to consider three dimensions and to proxy them using a small number of indicators. The deprivation cutoffs were based on those used by the UNDP (2014) or by the National Statistical Office, DANE, to compute the Basic Unmet Needs indicator. Moreover, all dimensions were given equal importance; in other words, each was assigned a weight of one third (1/3).

The Living Standards dimension considers access to Electricity, Water, Sanitation, Household Assets (i.e. electrical appliances), and Housing Materials. The Health dimension considers access to Health and Permanent Disability Insurance for all household members. Finally, the Education dimension includes School Attainment by adult household members and School Attendance by children belonging to the household. For more details on each indicator and its cutoff (namely, its deprivation thresholds) please refer to the methodological Appendix A1.

After identifying the deprived households identified by each indicator, a deprivation share (or score) was computed for each one. Next, in order to identify the multidimensionally poor, a multidimensional cutoff was chosen, which was used to filter households according to their deprivation share. Any household with a deprivation share above the one given by the multidimensional cutoff was regarded as poor. Finally, the MPI for Antioquia’s municipalities was computed. The MPI or -to use the term coined by Alkire and Foster (2011)- the M0, is the product of the average deprivation share and the headcount ratio.[4] The headcount ratio is the share of households within each municipality regarded as multidimensionally poor. The average deprivation share is the deprivation share of the typical household, in other words the average of the deprivation share.

Finally, in order to assess the robustness of the MPI in relation to the multidimensional cutoff, we computed the MPI using different cutoff values. The resulting orderings were then compared using three correlation measures: standard Pearson correlation coefficients, Kendall rank correlations, and Spearman rank correlations.

To understand the relationship between poverty and armed conflict in Antioquia, the study employed exploratory data analysis tools. First, we displayed the spatial distribution of the MPI and violence associated with the armed conflict using choropleth maps. In particular, we employed box maps to portray the spatial distribution of the variables of interest, classifying the sample into six groups, identical with the ranges that would have been portrayed in a standard boxplot. As the box map groupings were the same as those used in a boxplot, it was easy to spot spatial outliers in each one. A comparison of the resulting spatial patterns for the variables of interest might suggest the existence of spatial relationships in the data analysed.

Second, a set of kernel distributions was estimated, portraying the empirical distribution function of the MPI in the absence and presence of illegal armed groups. This allowed us to determine whether the distribution of poverty differs under the presence or absence of such groups.

Figure 6a presents the total number of armed attacks by illegal groups for the period 1996-2010, showing that the zones most affected by attacks by illegal groups were located in Antioquia’s north and southeast. Figure 6b also shows that the FARC guerrilla maintained a significant presence in the north during the period, including in the Urabá region, as well as in the west and southeast of the Department. The ELN guerrilla organisation also had a presence in the southeast, where at times they entered into conflict with the FARC. However, the ELN’s principal presence was in the northeastern region, where they were the dominant force. Finally, Figure 6d shows that the number of armed attacks by paramilitary groups was very low. This aspect has two potential explanations: 1) it is likely that there is an underreporting of paramilitary activities, as it is well known that these forces cooperated tacitly with the security forces in their offensive against guerrillas (Dube and Vargas, 2013; Acemoglu, Robinson, & Santos-Villagran, 2013), and 2) paramilitary forces used quite different strategies from those of the guerrilla groups in their quest for territorial control. In subsequent paragraphs, we will elaborate on these hypotheses further.

Figure 6
Number of Armed Actions, 1996-2010
Authors’ calculations with data from CINEP

A comparison of Figure 6 and Figure 3 suggests a fair amount of coincidence between areas with high levels of multidimensional poverty, and those hit hardest by Colombia’s armed conflict. Moreover, a comparison of Figures 6 and 4 also suggests that illegal groups prefer to locate in less accessible areas. However, against this general trend, Medellín and some municipalities in the east should be given special treatment: although they exhibit low levels of poverty, they were also the scenario of armed actions committed by different groups. Also bucking the general trend, the northernmost municipalities in Urabá, which are among the poorest, have experienced a low incidence of armed attacks.

Figure 7 illustrates an additional aspect of the armed conflict, showing the expulsion and reception rates of internal forced migrants. Specifically, it shows maximum municipal expulsion and reception rates between 1996 and 2010. It is fairly clear that the expulsion rate of forced migrants in the Medellín Metropolitan Area has been relatively low (Figure 7a). Indeed, Medellín has principally been a receptor of forced migrants (Figure 7b). The areas of Antioquia most heavily affected by armed conflict and that generated most forced displacement are the far western region, the southeast, and Urabá in the north. According to Figure 3, all these areas show a high MPI.

Figure 7
Forced Displacement: Maximum Expulsion and Reception Rates (x 100,000 inhabitants): 1996-2010
Authors’ calculations with data from CINEP

Counteracting these general trends, at this point it is worth noting 1) that Urabá’s northernmost municipalities are among the poorest in Antioquia, whereas Apartadó is the wealthiest city in region; 2) that despite this, and as is apparent from Figures 6a and 6b, Urabá’s northernmost municipalities experienced few armed attacks compared to other municipalities in Antioquia and in the region, though Apartadó itself was a frequent target of armed groups; and 3) that Urabá’s northern municipalities showed high rates of expulsion of forced migrants, whereas Apartadó was a net receiver.

This finding is consistent with Gómez, Sánchez-Ayala and Vargas, (2015), who show that forced displacement has often occurred in the absence of clashes or attacks by armed groups. These facts could be interpreted as a sign of the strategies used by paramilitary groups in their strongholds. Although prior to 1995 Urabá was under the control of the FARC, it later came to be dominated by paramilitary groups, who made the region one of their strongholds. It is well documented that paramilitary groups emerged in Urabá under the auspices of landlords –particularly cattle ranchers but also banana growers- and drug lords, and that they colluded with the state’s armed forces to expel the FARC (Dube and Naidu, 2010). This meant that, after securing control over Urabá, the paramilitaries granted themselves a virtual monopoly over violence in the region, acting with the support and complicity of a significant proportion of the local business and political elites. They then took advantage of their position to grab land by forcibly displacing the population, with the complicity of corrupt officials and notaries (Gómez et al., 2015). The land they seized was subsequently sold, or used for the pursuit of legal agribusiness activities. In this way, the illegal activities of the paramilitary groups became intertwined with legal activities, all under the auspices of local elites and state officials.

A more prototypical example of armed conflict is provided by Antioquia’s southeastern region (the Suroriente, see Appendix A1). Figure 3 shows that this region has suffered low to intermediate levels of multidimensional conflict. Simultaneously, Figures 6 and 7 show that this region experienced frequent armed attacks by ELN and FARC guerrillas, and was also the scenario of an important outflow of forced migrants. It may then be said that in the Suroriente the armed conflict manifested as a collision between two opposing groups for the control of the illegal rents in the region, which derived principally from extortion and kidnappings (Revista Semana, 2007).

As a method for collecting further evidence about the possible relationship between poverty and the armed conflict in Antioquia, the MPI distribution is conditioned by the presence or absence of illegal groups. That is, the conditional density function of the MPI is estimated. In Figure 8 the dashed lines represent the hypothetical distribution of the given MPI in the absence of illegal groups, while the solid line shows the opposite situation. It is apparent that the distribution of the MPI when illegal groups are present is flat and wide, with its distribution mass -i.e. its peak- located to the right of the distribution mass of the dashed line. Hence, we can hypothesise that armed groups tend to avoid the richest areas, in favour of slightly poorer ones. This might be explained by the unequal spatial distribution of state institutions and services, which in Colombia tend to be concentrated in the principal urban centres, which also happen to be wealthier than their surroundings. Therefore, notwithstanding the presence of armed gangs, it might be expected that better law enforcement in urban areas, such as the Medellín Metropolitan Area, would make it costlier for armed groups to operate openly in such places.

However, it should also be noted that the right tail of the solid line is thinner than that of the dashed line, leading us to hypothesise that armed groups prefer to avoid very poor areas, probably because the available resources are insufficient to generate looting opportunities or finance their activities. It seems, then, that armed groups prefer to operate in a species of middle ground, preferring municipalities that are not so poor that they do not have enough resources to finance illegal activities, but that at the same time are not so rich as to pose obstacles to their illegal activities in the form of law enforcement or other mechanisms.

Figure 8
MPI Distribution Conditioned On the Presence of Illegal Groups
Authors’ calculations with data from SISBEN and CINEP

Figure 9 shows the MPI distribution conditioned by the presence of FARC, the ELN, paramilitary counter-insurgency groups, and common criminality. The MPI kernel distribution conditioned by the presence or absence of the FARC guerrilla (figure 9a) is similar to the one described above. Taking the dashed line as a reference, the solid line indicating the presence of the FARC shows a flatter kernel distribution, which appears to shift towards the right. This suggests that the presence of this group tended to be associated with high poverty levels, in contrast with the situation prevailing in some of the municipalities in which the group did not establish itself. However, it is also apparent that the dashed line exhibits a thicker right tail. Thus, the FARC also avoided the extremely poor municipalities.

Regarding the presence of paramilitary counter-insurgency groups (figure 9c), it is found that the right tail of the solid line is thicker and gives a higher probability mass to the MPI values to the part of the dashed line furthest to the right, suggesting that the presence of the paramilitary may be related to extreme poverty. Thus, unlike the FARC, the paramilitary groups do not appear to have avoided extremely poor municipalities. At the same time, although it is found that the peak of the solid line is lower than the dashed one, both peaks are located on almost precisely the same point of the horizontal axis. Indeed, the left tail of the solid line is slightly thicker, suggesting that paramilitary groups, unlike the FARC, did not avoid the richest municipalities. This result is also consistent with the documented affinity between the Colombian security forces and the paramilitary counter-insurgency groups (Dube and Naidu, 2010), which permitted paramilitary groups to move freely, even in urban areas, and could help explain the underreporting of paramilitary-led attacks, particularly in Urabá and the Bajo Cauca and their core cities of Apartadó and Caucasia.

In the case of the ELN, Figure 9b shows that the right tail of the dashed line is thicker than the right tail of the solid line, allowing it to be said that the ELN has also tried to avoid extremely poor municipalities. However, as the solid line peak is slightly to the left of its dashed line counterpart, it appears that the ELN guerrilla is not afraid of assuming a presence in rich municipalities. This result could be related to the relative strength, historically, of the ELN in grassroots activity (Navarrete, 2014), which even gave them an edge in urban areas (Vélez, 2001), in contrast to the FARC, whose strategy was focused principally on seizing military control of vast tracts of territory and of small towns in rural areas. Also, it should be borne in mind that the ELN has, historically, been smaller and more nimble than the FARC. Therefore, the ELN developed an edge in using covert tactics that avoided direct clashes with the security forces, eschewed going for big targets, preferred to scatter its armed units, and relied heavily on civilian support (Vélez, 2001). In contrast, the FARC acted more like an army. In the words of one detained ELN commander -alias Byron- who fought unsuccessfully against the FARC in Antioquia’s Suroriente, “the FARC is an army of occupation” (“FARC contra ELN”, 2007). Indeed, at the height of its military power in the first decade of the 21st Century, the FARC was not afraid of directly defying the army.

These differences in tactics mean it was harder for the FARC to forge a presence in rich urban centres, whereas the grassroots tactics of the ELN granted it access to those centres without a need to exert complete territorial control. This implies that the ELN faced fewer obstacles to gaining a foothold in richer municipalities.

Finally, as Figure 9d shows, criminal activity is present across the entire range of the MPI, so that crime can be associated with both high and low levels of multidimensional poverty. However, compared with the guerrilla and paramilitary groups, criminal organisations appear to display a slight preference for rich, low poverty municipalities. This is probably consistent with the widespread presence of criminal gangs in urban centres.

Figure 9
MPI Distribution Conditioned by the Presence of the FARC, ELN, Paramilitaries, and Criminal Organisations
Authors’ calculations with data from SISBEN and CINEP