Forced Displacement in Colombia: Causality and Welfare Losses

Transcription

1 Forced Displacement Colombia: Causality and Welfare Losses Ana María Ibáñez Universad de los Andes Carrera 1E No.18A-10 Bogotá, Colombia Carlos Eduardo Vélez The World Bank/ LAC / PREM Room I8-113, 1818 H Street Washgton D.C., cevelez@worldbank.org June 2003 When this research was conducted, Ana María Ibáñez was a Research Associate at Fedesarrollo and Carlos Eduardo Vélez was a Senior Economist at the World Bank (LAC/PREM). The authors acknowledge and appreciate the permanent support and encouragement from Fernando Rojas (World Bank/LAC/PREM), manager of the grant (#<>) that proved fancial support to this paper. This paper does not necessarily reflect the views of the World Bank.

2 Forced Displacement Colombia: Causality and Welfare Losses Abstract Durg the last decade forced ternal displacement Colombia has been a growg phenomenon closely lked to the escalation of the ternal armed conflict - particularly rural areas. The displacement problem has hit nearly every region and vulnerable groups of the population. Two emergg policy questions are whether the magnitude of the response to this problem has been proportional to the size of the problem and to what extent the struments chosen are the most adequate to conta it. The purpose of this paper is twofold. First, to entify the determants of the displacement behavior and to compare those fdgs with standard migration literature. Second, to estimate the burden or welfare losses of displacement. Empirical evence shows that the welfare loss of displacement is conserable and amount to 25 percent of the net present value of rural life-time aggregate consumption, for the average divual. This loss is estimated for each divual with a method that derives welfare changes from behavioral model estimates wely used environmental and transport economics. Our empirical fdgs also show that the level of violence at the orig site is not only the domant factor of displacement behavior, but also that a violent environment other migration determants have the opposite effect, relative to the one expected by the migration literature a non-violent context. That is, the violent environment reverses the migration centives for risk aversion, access to formation, the planng horizon, location-specific assets human and non-human. Fally, separate modelg of preventive and reactive displacement behavior reveals higher welfare losses for more risk averse households and important asymmetries about the benefits of security force presence. 2

3 1. Introduction Forced displacement 1 Colombia is large and growg, covers nearly every region of the country and affects disproportionably vulnerable groups of the population. Durg the last fifteen years, voluntary displaced population is at least 1.8 million and corresponds to 4.3 percent of the Colombian population (World Bank, 2003). Intensification of the political conflict and its expansion to a vast majority of the territory is causg displacement numbers to grow at a larger pace than before. In 2001, 74 percent of Colombian municipalities were expulsion or reception sites. Mostly vulnerable groups compose the displaced population. Women, children and ethnic morities are respectively 49, 49 and 38 percent of this population (RSS,2002). In fact, the late 1990 s recent migrants (presumably, ternally displaced people) fared worse than the urban poor, clear contrast with the traditional migrant profile, who used to enjoy better welfare than the coverage urban population up to (see Vélez, 2002, Table 7.) This paper seeks to address two ma questions. First, it entifies the key determants of the displacement process. Understandg the determants of the process might shed some lights on possible policy struments to mitigate displacement. For example, it is not clear whether creasg protection of police or military forces the zones of potential conflict would reduce or crease displacement. Second, it estimates the burden of displacement monetary terms. The magnitude of welfare losses is relevant to justify policy terventions and vestments. Moreover, the size of public resources to alleviate displacement must take to conseration the extent of welfare losses duced by displacement. We fd that police and military forces assume differentiated roles preventg displacement. Police presence prevents displacement by reducg the likelihood of victimization but do not discourage displacement once families confront violence. On the other hand, military forces can protect the population once violence occurs and 1 The Inter-American Commission on Human Rights (1999) describes a displaced person as anyone who has been forced to migrate with the national boundaries, leavg ase her resence or her habitual economic activities because either her life, her physical tegrity or her freedom have been either violated or threatened by situations such as armed conflict, generalized violence, violation of human rights, and any other situation that may alter public order. 1

4 displacement is imment. Indeed, presence of military forces might become the only factor preventg displacement. Welfare losses from displacement are substantial. Compensatg valuation per household total average 25 percent of the net present value of rural aggregate consumption. Relative welfare losses are larger for the poorer segment of the displaced population. The paper is organized as follows. Section 2 describes some stylized facts about displacement Colombia and proves hypothesis on the possible causes of displacement. In Section 3, we present a brief literature review on migration literature, discuss its relevance for modelg the displacement decision and present a random utility model for displacement. Section 4 presents the empirical results and Section 5 concludes. 2. Displacement Colombia: Some Stylized Facts 2.1. Internally Displaced Population Colombia Forced displacement Colombia is the consequence of a political conflict between left-wg guerrilla, right wg paramilitary groups and the State. Escalation of crimes agast the civil population is a low cost and effective strategy to clear territories allowg illegal armed groups to strengthen their control area, transport weapons and develop at ease illegal activities (RSS, 2002). Displacement and political violence creased significantly durg the late neties. As the political conflict covers a sizeable proportion of the Colombian territory, most of Colombian municipalities face today displacement problems. Regional and household characteristics and the dynamic of the political conflict seem to determe displacement. This section describes some stylized facts about displacement Colombia. Violence and displacement are apparently strongly lked. War strategies adopted by illegal armed groups like death threats, massacres, forced recruitment, temporary town take-overs and selected homices force the civil population to flee their town. Figure 1 shows a likely correlation between crements of political homices and creases the total number of displaced households. In particular, durg the year 1999, the trend of political homices and displaced household soared significantly. Pérez (2002) found near 80 percent of violations to human rights and 82 percent of armed confrontations occur expulsion sites. 2

5 Displacement is a nationwe problem. The need of illegal armed groups to have territorial strongholds heightened and expanded the conflict across the country. As a consequence, near 74 percent of Colombian municipalities receive or force out population and with the exception of one department 2, an island the Caribbean Sea, all departments face displacement problems (Map1 and 2). Nevertheless, tensity of displacement 3 is heterogeneous across and with departments. While the department exhibitg the largest displacement tensity 10 percent of the population fled, the department with the fifth highest tensity over four percent left (Figure 2). This we variance across departments suggests regional characteristics might determe partially displacement behavior. For example, illegal armed groups may want to control territories rich natural resources. Particular characteristics of households may trigger displacement as well. Some socio-demographic factors and the particular social context where the household rese may crease the likelihood of beg victimized. On the other hand, some households may be more risk averse and may prefer to leave their town to prevent beg the victim of violence. Social context dimensions such as risk and protective factors paired with victimization are seemgly important determants of displacement. Presence of illegal armed appears to promote displacement. Figure 3 dicates displaced households use to rese regions where paramilitary and guerrilla presence is strong while military presence is weak. Facg violence, particular death threats to household members, pushes households to seek refuge elsewhere. Indirect violence, such as massacres nearby towns or murder of a friend, is as well a source of displacement, albeit death threats play a stronger role (Figure 4). Household characteristics may fluence the likelihood of death threats, victimization and displacement. Landowners are four times more cled to flee their hometown (Figure 5). Apparently, illegal armed groups are terested violently appropriatg land. Displaced households participated actively community activities their orig site contrast to household who d not displace (Figure 6). Illegal armed groups seek to destroy social cohesion conflict zones by timatg leaders and 2 Departments are equivalent to states. 3 Intensity of displacement is measured as the number of displaced population per habitants 3

6 active members of the community to elimate the chances of civil opposition. Other factors, such as access to social services, elevate migration costs and centives resents to stay their hometown (Figure 7). Constant changes the conflict dynamic caused modifications displacement typology. Durg the neties, households displaced maly after a town massacre and migration was massive, that is a group of households fled together simultaneously reactg to a violent act. Today, many causes trigger displacement and households relocate mostly divually to prevent victimization. From 2000 to 2001, divual displacements soared 414 percent stark contrast to an crease of 58 percent of massive displacements. Indivual relocation is closely related to preventive displacement while massive relocation is lked to preventive displacement (RSS, 2002). Preventive displacement may contue to rise as a consequence of the geographic extension of the conflict (Meertens, 1999) Alternative Explanations of Displacement Colombia Causes of voluntary migration Colombia are difficult to entify. Immediate causes or triggers are the last cent a cha of events that produce the fal decision to flee the hometown. And yet the root of displacement underlies the dynamics of the Colombian conflict. This section describes some hypothesis put forth the literature about the sources origatg displacement Colombia. Illegal armed groups and its actions agast the civil population are maly responsible for forced displacement. In 2001, paramilitary groups stigated 50 percent of displacements while guerrilla and two actors simultaneously origated 20 and 22 percent respectively (RSS, 2002). Paramilitary groups not only bear the bulk of the responsibility but also are more effective displacg population. Durg 2001, paramilitaries caused 599 displacement events that forced out people meanwhile guerrilla groups provoked 570 events that prompted people to flee (RSS, 2002). Violent actions agast the civil population, like threats and selective homices, trigger the decision to displace. Threats and homices are the ma motive to seek refuge elsewhere (Conferencia Episcopal, 1995, Arquiócesis de Bogotá - Codhes, 1997). However, armed confrontations became lately an important trigger due to the recent tensification of the conflict populated areas (RSS, 2002). 4

7 Land conflicts and violent land appropriation is consered an underlyg source of displacement (Reyes and Bejarano, 1998). Land occupation is crucial the war strategy to clear the territory from the presence of opponents, to expand control areas and to appropriate valuable land. A low cost strategy to occupy land is to drive out small landholders and appropriate their land (USCR, 2001). Displaced population reports to have lost four million hectares of land 4, which amounts to one third of productive land Colombia (PMA, 2001). Programs to eradicate illicit crops may as well produce displacement (Arquiócesis de Bogotá-Codhes, 1997). Aerial fumigation of illicit crops 5 destroys farmers assets, produces a temporal shock on their come and origates combats exacerbatg violence the region. Estimations dicate people displaced durg 1999 drug producg departments (Puyana, 1999). Forcg out population may be a war strategy to impede collective action, to damage social networks as well as to timate and control the civil population. Attacks to the populations weaken their support to the opponent and obstruct rise up of civil population (Henao et al., 1998). Lozano and Osorio (1999) estimate 65 percent of displaced population was an active member of community organizations and 11 percent participated labor and politic organizations their hometown. Rural families may voluntarily migrate to avo forced recruitment of their children. Children as young as eight years are currently recruited by illegal armed groups to fight as soldiers the Colombian conflict (Salazar, 2001). After a combat October 2001, military forces found 43 percent of dead guerrilla members and 41 percent of captured guerrilla members were below 18 years of age (USCR, 2001). 3. Modelg Displacement as Migration 3.1. Logical asymmetries between migration and displacement centives Somehow, displacement resembles migration behavior. Households must compare the benefits and costs from resg the orig and reception sites and choose the alternative with larger net benefits. Nevertheless, the case of displacement, 4 These figures might overestimate the total hectares of abandoned land because displaced households have centives to report ownership of larger farm sizes the event a program of land restitution is implemented. 5 Programs to eradicate illicit crops follow two strategies: (i) aerial fumigation of illicit crops; or (ii) manual and voluntary substitution. Some analysts conser aerial fumigation is causg displacement. 5

8 violence is an additional factor the decision process that modifies the costs from stayg at the orig site and, consequently, might modify the impact of other migration determants. The purpose of this section is to discuss the variables entified the literature as determants of the migration decision and to show, most cases, centives are reversed forced displacement. Reversal occurs because violence reduces returns and creases risk the site of orig. Determants of costs the destation site fluence the decision to migrate and to displace the same direction. Contacts at the reception site and education decrease migration costs (Becker, 1975; Todaro, 1989; Todaro and Maruszko, 1987). Contacts at the reception site may prove housg, support to fd employment and a social network. Better-educated divuals may fd employment easier and generate larger comes after migratg. On the other hand, potential discrimation at arrival creases migration costs and, thereby, discourages migration (Fischer et al. 1997). In Colombian urban centers, discrimation agast displaced population is particularly strong. Some native resents wrongly believe displaced households belong to illegal armed groups and, addition, perceive this population attracts public resources previously allocated for the poor. Empirical evence suggests that the long-standg conflict Colombia reverses the effect of some traditional migration determants. The length of the planng horizon exerts similar centives on the decision to migrate and to displace but the underlyg motives differ. Inclation to migrate is larger for divuals with long planng horizons (Becker, 1975; Todaro, 1989; Todaro and Maruszko, 1987). On the other hand, young people are probable targets of threats, forced recruitment and selective homices; therefore likelihood to displace is larger. Positive formation about economic and social opportunities the destation site improves the benefits from migration (Stark and Levhari, 1982; Dustmann, 1992; Maier, 1985). Conversely, formation about poor social and economic conditions destation sites raises the benefits of non-migration. Risk aversion assumes an asymmetric role the decision to migrate and to displace. The uncertaties herent arrivg to an unknown place may dissuade risk averse divuals to migrate (Fischer, 1997). In contrast, violence may duce risk averse 6

9 households to displace spite of the complications they might cope with the reception site. Bounded rationality is a crucial component to conser displacement analysis. Decisions are often limited by past experiences, emotional patterns and complexities of evaluatg benefits and costs (Simon, 1983). The Colombian confrontation pushes divuals to extreme situations that hder their capacity to dece rationally. Displaced households decg under fear may opt for sub-optimal options because they underestimate displacement costs, overestimate risks and utilize high discount rates. Location specific assets render migration costly and reduce centives to migrate (Fischer et al, 1997). Deficient rule of law Colombia leaves unprotected location specific assets allowg illegal armed groups to violently appropriate land. Landownership becomes, under these special circumstances, a possible factor of victimization and a cause of displacement. Similarly, human capital location specific assets play an asymmetric role when destruction of social networks is a war strategy. Permanent resency and active participation community activities signifies advantages from belongg to a society. Discouragg migration to the extent that it would entail givg up these accumulated advantages (Fischer et al., 1997). Sce destruction of social networks is a war strategy, high levels of social capital is no longer an asset but a risk factor. Results from the migration literature are modified when violence enters to the decision process. Contacts at reception site, education and discrimation at arrival determe migration and displacement the same direction. On the other hand, violence reverses the effect on the migration decision of the planng horizon, access to formation, risk aversion, bounded rationality and location specific assets 7

10 3.2. A Random Utility Model for Displacement 6 Household i deces whether to displace if the utility from displacement is greater than the utility from stayg the orig site U > U. (1) where U ij denotes the direct utility from alternative j, j=d is the reception site and j=n is the orig site. The direct utility is composed by the determistic utility ( v ij ) and a random term ( ε ij ) with mean zero U = ij v ij + ε. ij (2) Decision to displace or rema the orig site depends on many factors. First, households evaluate risks and generate expectations about security the orig and destation region ( S ij ). Second, households compare come possibilities and access to social services both sites ( Y ij ). Third, migration and formation costs fluence the decision process ( C ). Lastly, household characteristics reflectg preference on ij needs and risk aversion determe displacement behavior ( Z ). The observable utility is defed as v = ij f ( S Y, C Z ). ij, ij ij i If we assume a logistic distribution for the error term and a lear utility function, the probability of displacement is (3) i prob i ( displace) exp = 1+ exp ( α( S S ) + β dy β ny + δ ( C C ) + ( γ γ ) Z i ) ( α( S S ) + β Y β Y + δ ( C C ) + ( γ γ ) Z ) d n i. (4) Equation (4) assumes margal utility of come changes after displacement. Perceptions of security can be approximated with variables dicatg whether the household was directly threatened and whether the household is facg direct violence. Direct threats are however endogenous. Landowners, active members of 6 This model was developed Kirchhoff and Ibáñez (2001). 8

11 the community or young household heads are possible targets of illegal armed groups. The predicted threat of household i is equal to Pr ob ( Threat) = i f ( L, V, A Z i ), (5) where L denotes landownership the place of orig, V is ties the place of orig and A is the presence of armed actors the place of orig. The fitted value of the predicted threat will be cluded as an exogenous variable the probability of displacement. Displaced households confront welfare losses from the deterioration of their quality of life. These losses, although not manifested monetary terms, are likely to be one of the most significant costs of displacement for Colombian society. If these costs to the displaced themselves are left out evaluatg the dimension of the problem, the policies implemented to alleviate displacement might be sufficient or misgued. We will estimate welfare losses from displacement based on methods used wely environmental economics. The shock from displacement exhibits a similar structure than environmental problems. An external shock, this case violence, duces changes behavior, which turn imposes welfare losses to households. One way of measurg changes utility monetary units is compensatg variation. Compensatg variation for avog displacement is the amount of money necessary to leave the divual different between displacg and stayg his hometown. In this case, compensatg variation can be terpreted as a measure of the willgness to accept come exchange for not displacg. As shown by Hanemann (1982), compensatg variation (CV) can be defed as the measure that equates the expected maximum utility before and after the displacement. For the model explaed above, expected compensatg variation can be defed as 7 [ ] E CV i α ( S S ) + β dy β ny + δ ( C C ) + ( γ γ ) Z = β The theoretical contributions of the model defed above are twofold. First, the random utility model permits to troduce variables never consered migration models, such as perceptions of security, and gives enough flexibility for clusion of d i 7 A complete derivation of the compensatg variation is derived Appendix I. 9

12 reverse centives. Second, the defition of welfare losses allows policy makers to dece whether tervention is necessary and establishes an upper bound for vestment funds to mitigate displacement. The random utility model defe above, typically used environmental and transport economics, allows us to retrieve the parameters of the utility function and, thereby, to estimate welfare losses. 4. Determants of Displacement Colombia and Associated Welfare Losses 4.1. The Data The purpose of the Survey for Internally Displaced Population 8 (SIDP-2000) was to entify the causes of displacement Colombia and to measure its associated welfare losses. Surveys were conducted orig and destation sites order to have formation about displaced households and households who d not displace despite leavg conflict zones - hereafter non-displaced households. Two samples were constructed: displaced and non-displaced sample. The questionnaires that were admistered to these households covered issues that ranged from socio-economic characteristics of the household, victimization profile, armed actors the region, access to social services the orig and destation site, land ownership and agricultural production. The sample for displaced households was selected destation sites with the largest fluxes of displaced population durg The surveys were admistered to 200 displaced households Bogotá, Cartagena and Medellín. Questionnaires were applied only to households displacg from Antioquia and Cordoba, the departments with the highest records of population expulsion The regional composition of the displaced sample was purposively chosen with the objective of buildg a counterfactual sample of non-displaced population with a similar regional composition. The non-displaced sample was composed of 176 surveys of households resg conflict zones traditionally affected by displacement and located Antioquia and Cordoba. Although the survey proves valuable formation about forced displacement Colombia, the sample was not representative of the displaced population therefore results cannot be generalized. 8 Detailed description of the survey can be found Kirchhoff and Ibáñez (2001). 10

13 Table 1 presents summary statistics for exogenous variables for the displaced and non-displaced sample. The descriptive statistics proves some itial sights on displacement behavior. First, displaced and non-displaced households are exposed to excessive violence levels. Near 50 per cent of displaced households and 24 per cent of non-displaced households faced direct threats the orig site. Moreover, few households have not confronted direct violence 9 : 94 per cent of displaced households and 77 per cent of non-displaced households reported beg victim of direct violence. Second, non-displaced households feel better protected by government forces. In contrast to non-displaced households, displaced households perceive a greater presence of paramilitary and guerrilla their hometown and a weaker presence of police and military forces. Third, evence suggests violence is not randomly targeted. Displaced households are landowners larger proportions, participate more organizations and have younger household heads than nondisplaced households. Yet land size is larger for non-displaced households, which may imply illegal armed groups mostly target landowners with small farms. Lastly, non-displaced apparently have a higher economic status because they are better educated, have more accesses to basic social services 10 consumption 11 is larger when compared to displaced households. Table 1. Descriptive Statistics Displaced Non Displaced Mean Variance Mean Variance Direct threat Indirect violence Paramilitary presence Guerilla presence Military presence Police presence Contacts reception site Years of resence orig site Own land Standardized land size Access to social services Household education Access to media Rural consumption a Urban consumption a Age household head Household size before displacement and rural aggregate 9 A household was defed to confront direct violence when a nearby town or when friends and family were the victims of attacks by illegal armed groups, massacres, bombs or any other type of violence. 10 Access to basic social services is a dummy variable equal to one when the household has access to education and health. 11 Appendix I describes the methodology used to predict rural and urban aggregate consumption. 11

14 Male household head Number of organizations Source: Authors calculations based on SIDP-2000 a. In million pesos 4.2. Predicted Threat Model Aggressions agast the civil population are not randomly targeted. Last sections prove evence that illustrates illegal armed groups may attack households with particular characteristics. Direct threats are consequently an endogenous variable. In order to reduce endogeneity problems, we estimate the probability of a household beg the victim of direct threats usg a probit model. Fitted values of the estimation are cluded the probability of displacement as an endogenous variable. Table 2 reports the results for the predicted threat model. The most likely victims of direct threats are landowners as well as families with young household heads and with household members actively volved community activities. These results confirm the hypotheses developed the literature. Illegal armed groups are violently appropriatg land and attack young and active members of the community as part of a war strategy. Estimations dicate as well households resg zones of paramilitary domance are threatened with a larger probability whereas guerrilla domance does not seem to have a significant effect on threats. This result should be carefully analyzed. When the SIDP-2000 survey was conducted, displacement occurred maly as a consequence of paramilitary actions like threats and massacres. Nevertheless, the dynamics of the conflict changed significantly durg the last years and today guerrilla groups are responsible for many displacement events. Lately, guerrilla attacks to small and medium municipalities have provoked large expulsions of population 12. Police protection deters threats from illegal armed groups to the population and, consequently, prevents displacement. Contrary, military forces are not strumental to reduce the likelihood of threats. Protection of the civil population requires a 12 For example, May 2002 leftist guerrilla groups attacked Bojayá, a small municipality located on the Pacific Ocean. As a result of the attack, 119 people died and people were forced to displace (CE, 2002). 12

15 constant presence of the State and a reliable stitution with strong lks with the community. Police forces embody these conditions. The role of military forces is, on the other hand, to protect the population durg war. Yet their presence should not be permanent each Colombian municipality. Table 2. Probability of Threats Margal P> z Effect Own land Years of resence orig site Age household head Number of organizations Paramilitary presence Guerrilla presence Military presence Police presence Number of observations 363 Pseudo R-square Source: Authors calculations based on SIDP Determants of Displacement The displacement model defed section III is estimated usg maximum likelihood procedures. Three models are estimated. The first is the Aggregated Model that makes no distction between preventive and reactive displacement. In the next section, we estimate a model distguishg by displacement types preventive and reactive. Table 3 reports estimation results for the Aggregated Model. Security perceptions are the domant predictor of displacement. In particular, an crease of one per cent predicted threats 13 raises the probability of displacement by 326 per cent. Indirect violence, although not as effective as predicted threats, is also a domant factor of displacement. Military protection discourages displacement once threats and direct violence has been taken to account. The role of police presence is not significant when displacement is imment. Displacement costs, though significant, do not counterbalance the effects of violence the orig site. Households with access to basic social services are less likely to displace. Access to media, probably by provg formation about difficulties families face reception sites, dissuades displacement. However, the jot effect of both variables is not enough to compensate the fluence of direct violence let alone 13 Direct threats are endogenous to the model and were strumented usg predicted threats. The model meets the entification conditions. 13

16 predicted threats. As expected, contacts availability at reception creases likelihood of displacement. Consumption dicators the displacement decision behave similarly than migration models. Foregone consumption the orig site decreases the chances of displacement while consumption opportunities the destation site centive displacement. Unlike migration models, better-educated household are less willg to displace. Probably better-off households are able to adopt protective measures. Evence from descriptive statistics reported Table 1 supports these fdgs because non-displaced are better-educated households with larger farm sizes and higher rural consumption. Household characteristics determe partially the decision to displace. Households with younger heads are more cled to displace. As previously discussed, young divuals are likely targets of illegal armed groups but the predicted threat already accounts for this effect. The tendency of younger heads to migrate may reflect risk preferences of households. On the other hand, years of resence crease the probability of displacement. Table 3. Probability of Displacement Aggregated Model Margal Effect P> z Predicted threat Indirect violence Military presence Police presence Contacts reception site Years of resence orig site Standardized land size Access to social services Household education Access to media Rural consumption a Urban consumption a Age household head Male household head Number of observations 361 Pseudo R-square Source: Authors calculations based on SIDP a. In million pesos The model estimated above is robust to changes specification. We estimated various specifications of the model and results d not vary significantly. Parameter 14

17 estimates, t-statistics and pseudo r-squares are practically entical when other models are estimated. Empirical estimation confirms the reversal of migration centives when life threats, lack of rule of law and violation of property rights prevail. Violence and aggressions agast the civil population modifies the migration centives of education and location specific assets, such as land and social capital. Other migration determants, like consumption dicators and access to basic social services, fluence displacement decisions the expected direction. Military and police protection reduce displacement but different stages of the process. Police protection is paramount to ease aggressions of illegal armed groups to the civil population. Once the civil population faces aggressions, military protection is the only possible strument to halt displacement Modelg Two Displacement Types: Preventive and Reactive The previous model is estimated for preventive and reactive displacement. We defed preventive displacement when households entified fear despite not beg threatened as reason for fleeg their hometown. Results for the preventive and reactive model are presented Table 4. Perceptions of security and some migration variables are similar the preventive and reactive displacement models. Predicted threats and direct violence contue to be the ma determants of displacement. Both variables are however stronger reactive displacement. Ak to the Aggregated Model better-educated and older household heads are less likely to displace the preventive and reactive model. Urban consumption promotes displacement both models. Results suggest the more risk averse self-select to preventive displacement. First, military protection is strongly significant for reactive displacement but seems sufficient for preventive households who prefer to leave before violence escalates. Second, rural consumption is irrelevant for preventive displacement. Risk associated with rural consumption is weakly compensated with creases the mean of rural consumption. Unlike reactive displacement, migration costs contribute to promote or deter preventive displacement. Contacts at destation site duce households to migrate 15

18 only for preventive displacement. Access to social services remas significant for both displacement types but a bit stronger preventive displacement. Table 4. Probability of Displacement Preventive and Reactive Displacement Preventive Displacement Reactive Displacement Margal P> z Margal P> z Effect Effect Predicted threat Indirect violence Military presence Police presence Contacts reception site Years of resence orig site Standardized land size Access to social services Household education Access to media Per capita rural consumption a Per capita urban consumption a Age household head Male household head Number of observations Pseudo R-square Source: Authors calculations based on SIDP-2000 a. In million pesos Empirical fdgs show violence, especially direct threats and direct violence, rema the domant factor of both types of displacement. Yet behavior of preventive and reactive types is partially different. Immediate response of military forces can reduce reactive displacement while preventive displacement is difficult to halt. Military protection does not dissuade preventive displacement. On the other, police protection, by reducg the probability of threats, mitigates preventive displacement. Access to social services, although statistically significant, does not prove a clear centive to deter households from displacg. Indeed, contag violence is the only effective strument to control preventive displacement Welfare Losses Welfare losses from displacement are substantial. In fact, the costs from displacement amount to 25 per cent of the net present value of aggregated rural consumption (See Table 5). Figure 8 shows the cumulative distribution of welfare losses as a percentage of net present value of rural aggregate consumption. Near 40 percent of households experience welfare losses above 40 percent of the net present value of the aggregated rural consumption. 16

19 Table 5. Welfare Losses as Percentage of Rural Aggregate Consumption % Welfare Losses Aggregated Model 25% Preventive Displacement 24% Reactive Displacement 15% Source: Authors calculations based on SIDP-2000 When welfare losses are estimated for preventive and reactive displacement, we fd preventive displacement generates larger welfare losses, 24 per cent, contrast to reactive displacement, 15 per cent (Table 5). Risk aversion might determe this gap because welfare losses from preventive displaced population may encompass regular economic costs from displacement as well as costs from facg uncertaty and anxiety. The economic burden of displacement falls disproportionably upon the poor. Figure 9 plots welfare losses as a percentage of rural aggregate consumption agast rural aggregate consumption per capita. Poorer households suffer losses above 60 percent while this figure ranges from 20 to 40 percent for families with larger consumptions. Unfortunately, the economic literature does not prove similar estimations to compare the size of welfare losses from displacement Colombia. However, we can contrast our results to estimation of welfare losses from environmental degradation other countries. For example, welfare costs from air pollution Bogotá, the fourth most polluted city Lat America, total 0.01 per cent of the net present value of household come 14 (Ibánez and McConnell, 2001). Economic costs from air pollution Taiwan are equal to per cent of the net present value of household come 15 (Alberi et al, 1997). 5. Conclusions Forced displacement modelg diverges from traditional migration modelg. Many key determants of migration have the opposite effect the context of forced displacement. Our empirical fdgs confirm this hypothesis. Violence the orig 14 This measure estimates willgness to pay for reductions one symptom day of acute respiratory illnesses. 15 This measure estimates willgness to pay to avo recurrence of an episode of acute respiratory illnesses. 17

20 site modifies the migration centives of education and location specific assets, such as land and social capital. Large welfare losses justify policy tervention. Economic costs of displacement are average 25 per cent of the net present value of aggregated rural consumption. Moreover, poorer families experience larger welfare losses. In fact, some households have welfare losses above 80 percent of the net present value of aggregated rural consumption. Our estimations prove evence on possible policy struments to prevent displacement. Violence, particular direct threat and direct violence, is the major determant of displacement and is, thereby, the key strument to prevent displacement. Other type of terventions has a margal effect on displacement and cannot compensate the effect of direct threats and direct violence. However, police and military protection can mitigate displacement. While police presence prevents direct threats, military forces are strumental to protect the population once displacement is imment. On the other hand, economic variables, like access to basic social services or access to formation, mildly prevent displacement. 18

21 References Alberi, A., M. Cropper, T. Fu, A. Krupnick, J. Liu, D. Shaw, and W. Harrgton. Valug Health Effects of Air Pollution Developg Countries: The Case of Taiwan, Journal of Environmental Economics and Management, 34: (1997). Arquiócesis de Bogota-CODHES, (1997). Desplazados por violencia y conflicto social en Bogotá. Bogotá, Colombia. Becker, G. (1975) Human Capital, Columbia University Press, New York.. Conferencia Episcopal (1995). Derechos Humanos - Desplazados por Violencia en Colombia. Bogotá, Colombia. Conferencia Episcopal (2002). RUT forma sobre desplazamiento forzoso. Estudio de Caso 4. Bogotá, Colombia. Dustmann, C. (1992) Migration, Savgs and Uncertaty, Department of Economics, European University Institute, Firenze, Italy. Fischer, P.A. et al. (1997) Should I Stay Or Should I Go?, International Migration, Immobility and Development (ed. Hammar, T.), Berg, Oxford. Hanemann, M.W. (1982) Applied Welfare Analysis with Qualitative Response Models. California Agricultural Experiment Station Workg Paper No Berkeley, CA: University of California. Hanemann, M.W. (1984). Welfare Evaluations Contgent Valuation Experiments with Discrete Responses, American Journal of Agricultural Economics, 66(3): Henao, H. et al. (1998). Desarraigo y futuro. Va cotiana de familias deslazadas de Urabá. Medellín, Colombia. Ibáñez, A.M. and T. McConnell (2001). Valug Morbity: Acute Respiratory Illnesses Bogotá, Colombia. Unpublished Mimeo. Interamerican Commission on Human Rights (1999). Desplazamiento Forzado Interno. Chapter VI of Tercer forme sobre la situación de los derechos humanos en Colombia. Organization of American States, Washgton D.C. Kirchhoff, S. and Ibáñez, A.M. (2001). Displacement Due to Violence Colombia: Determants and Consequences at the Household Level, ZEF Discussion Papers on Development Policy No. 41. Bonn University. Lozano, F.A. y F.E. Osorio (1999). Horizontes de comprensión y acción sobre el desplazamiento de población rural en Colombia ( ). CODHES. Bogotá, Colombia. Maier, G. (1985) Cumulative Causation and Selectivity Labour Market Oriented Migration Caused by Imperfect Information, Regional Studies, vol. 19, pp Meertens, D. (1999) Desplazamiento forzado y género: trayectorias y estrategias de reconstrucción vital Desplazados, Migraciones Internas y Reestructuraciones Territoriales (eds. Cubes, F. y C. Domínguez). Centro de Estudios Sociales Universad Nacional y Misterio del Interior. Bogotá, Colombia. Pérez, L.E. (2002). Desplazamiento forzado en Colombia : Una aproximación empírica a las relaciones entre desplazamiento, conflicto armado y desarrollo en El desplazamiento forzado en Colombia: compromisos desde la universad. Bogotá, Colombia. 19

22 PMA, Programa Mundial de Alimentos (2001). Estudio de caso de las necesades alimentarias de la población desplazada en Colombia. Bogotá, Colombia. Puyana, A.M. (1999). Cultivos Ilícitos, fumigación y desplazamiento en la Amazonía y la Oroquía Desplazados, Migraciones Internas y Reestructuraciones Territoriales (eds. Cubes, F. y C. Domínguez). Centro de Estudios Sociales Universad Nacional y Misterio del Interior. Bogotá, Colombia. Reyes, A. y A.M. Bejarano (1998). Conflictos agrarios y luchas armadas en la Colombia contemporánea. Análisis Político 5:6-27. RSS, Red de Solarad (2002). Informe al Congreso de la República. Presencia de la República Enero Febrero Bogotá, Colombia Salazar, M. C. (2001). "Consequences of armed conflict and ternal displacement for children Colombia," Wnipeg Conference on War Affected Children, Report of the United Nations High Commissioner for Human Rights on the Human Rights Situation Colombia. Simon, H.A. (1983) Reason Human Affairs, Basil Blackwell, Oxford. Stark, O. and D. Levhari (1982) On Migration and Risk LDCs, Economic Development and Cultural Change, vol. 31, pp Todaro, M.P. (1989) Economic Development the Third World, Longman, New York. Todaro, M.P. and L. Maruszko (1987) Illegal Immigration and US Immigration Reform: A Conceptual Framework, Population and Development Review, vol. 13, pp USCR U.S. Committee for Refugees (2002). Report Washigton. D.C. Vélez, C.E. (2002). Colombia Poverty Report. World Bank: Washgton, DC. World Bank Colombia: The Economic Foundation of Peace. Washgton. D.C. Wodon, Q.T. (1999). The Micro Determants of Consumption, Poverty, Growth and Inequality Bangladesh. World Bank Policy Research Workg Paper Washgton, DC. 20

23 APPENDIX I Derivation of Compensatg Variation 16 The utility from displacement for household i is defed as U = α S + β Y + δc + γ Z + ε. d i On the other hand, the utility for household i from resg the orig site is U = α S + β Y + δc + γ Z + ε. n i The money value necessary to equate the utility before and after displacement, that is the compensatg variation, is equivalent to d ( Y CVi ) + δc + γ Z i + ε = αs + β ny + δc + γ Z i ε α S + β + which becomes CV i α( S = S ) + β Y d β Y n + δ ( C β d C ) + ( γ γ ) Z Sce ε and ε are random variables with mean zero, the expected compensatg variation is defed as [ ] E CV i α ( S S ) + β Y β Y + δ ( C C ) + ( γ γ ) Z d i + ε d n i =. β ε,. 16 The derivation of the compensatg variation draws on Hanemann (1984). 1

24 APPENDIX II Prediction of consumption aggregate To estimate the consumption aggregate of SIDP-2000 households, we estimated a regression for the micro determants of consumption for urban and rural areas utilizg the Encuesta de Calad de Va (1997). The coefficients from the estimation were used to predict urban and rural consumption for displaced households. Based on Wooodon (1999) and the results for Vélez (2002), we cluded the followg determants of consumption cluded: (i) regional controls; (ii) household size variables: the number of babies, children and adults; (iii) other demographic and gender variables such as gender and age of household head as well as family structure; (iv) education variables: education of the household head and education of the spouse; (v) the standardized amount of land owned for rural areas; and (vi) rural migration status for urban areas. Results for the urban and rural estimation are presented Tables II.1 and II.2. Table II.1. Estimate for rural consumption Variable Coefficient P> z Number of children under 2 years -84,0 0,73 Number of children under 2 years squared 46,2 0,78 Number of children between 3 and 13 years -32,4 0,68 Number of children between 3 and 13 years squared 8.894,2 0,56 Number of adults (14-65) 371,3 0,00 Number of adults (14-65) squared 8.818,0 0,95 Age household head 57,8 0,00 Age household head squared ,5 0,00 Male household head 533,5 0,00 Years of education household head 103,7 0,00 Years of education household head squared 3.269,7 0,25 Years of education spouse 112,6 0,00 Years of education spouse squared 6,6 0,01 No spouse -312,9 0,04 Constant 192,3 0,69 Adjusted R-Square F Test Source: Authors calculation based on Encuesta de Calad de Va (1997) *Regional controls cluded 2

25 Table II.2 Estimate for urban consumption Variable Coefficient P> z Number of children under 2 years ,1 0,10 Number of children under 2 years squared 608,8 0,36 Number of children between 3 and 13 years -450,9 0,08 Number of children between 3 and 13 years squared 130,4 0,05 Number of adults (14-65) 654,6 0,01 Number of adults (14-65) squared 2.494,9 0,95 Age household head 276,3 0,00 Age household head squared -2,2 0,00 Male household head 1.018,2 0,02 Years of education household head -214,2 0,03 Years of education household head squared 35,0 0,00 Years of education spouse 218,7 0,02 Years of education spouse squared 10,1 0,04 No spouse ,4 0,01 Rural migrant -374,5 0,28 Constant ,6 0,00 Adjusted R-square F Test Source: Authors calculation based on Encuesta de Calad de Va (1997) *Regional controls cluded 3

26 Figure 1. Displaced Households and Violence Displaced households Political homices Displaced households Homices Source: RSS and Human Rights Commission 1

27 Source: RSS 1

28 Source: RSS 1

29 Figure 2. Displacement Intensity by Department 2002 Bolívar Sucre Caquetá Source: RSS Putumayo Chocó Source: Rss (2002) Displaced population/100,000 habitants 1

30 Figure 3. Perception - Presence of Armed Groups 100% Percentage of households 80% 60% 40% 20% 0% Military presence Paramilitary presence Guerrilla presence Displaced Non Displaced Source: SIDP

31 Figure 4. Direct threats and Indirect Violence 100% Percentage of households 80% 60% 40% 20% 0% Direct threats Indirect violence Displaced Non Displaced 1

32 Figure 5. Land Ownership 50% Percentage of households 40% 30% 20% 10% 0% Displaced Non Displaced Source: SIDP

33 Figure 6. Number of organizations Number of organizations per household Displaced Non Displaced Source: SIDP

34 Figure 7. Access to Social Services 100% Percentage of households 80% 60% 40% 20% 0% Displaced Non Displaced Source: SIDP