Determinants of Migration Patterns of Colombia using the Students Data : from where and to where Colombian migrants move

Transcription

1 Determinants of Migration Patterns of Colombia using the Students Data : from where and to where Colombian migrants move Abstract This study analyses the migration patterns of Colombia taking into account poverty and violence variables. The data used in this research are the Ministry of Education records of enrolments from 2006 to 2009, which include all students in official schools (72% of students in Colombia). Although the data is restricted to students between 5 and 17 years, it offers a very clear picture of internal migration in Colombia and its results coincide with the national trends described by other authors. The methodology used was implemented by the Office of the Deputy Prime Minister of the United Kingdom in 2002 to analyse spatial interaction in this country. Four of the most important results are: 1) in Colombia, migration does not imply an improvement in life conditions, since most of students move to lower socioeconomic stratums regardless of the zone (urban or rural) of their origin or destination; 2) the effect of wealth at municipal level in Colombia agrees with the theory of the migration hump; 3) the effects of violence and unemployment on out-migration are conditioned to the wealth of the municipality and 4) migration flows from municipalities with armed conflict and/or illicit crops are addressed to municipalities with worse living conditions in most cases. Introduction This is the first chapter of a study whose aim is to determine the incidence of inter-municipal migrations and changes in family structure on school achievement of students, emphasizing the effect of poverty and armed conflict in Colombia. Thus this chapter analyses the migration patterns of Colombia taking into account poverty and violence variables. The second chapter studies the effect of these variables on household formation and its reorganization after migrating, using Sisben data 1. Finally, the last chapter combines the 1 Sisbén is an information system classifying the population in accordance with their social and economic status in order to identify the possible beneficiaries of subsides granted by the Government and the Municipality to the poorest population and vulnerable in: education, housing, health, as well as to have access to the Certificate of Military Service, subsides to the aged and basic food assistance. Sisben s survey includes questions on education, housing, social security and demographic conditions at home. 1

2 results of the two previous chapters to study the school achievement of migrant and displaced students. The results from this study are very useful for the design of education and child development programs as well as policies for poverty reduction. For instance, one of the policies that could impel this study is the creation of a program to even knowledge from migrant students, before entering to a regular school, with the knowledge of those students from the receiving municipality. This would imply an additional investment of resources; however, this program could improve the efficiency of educational institutions, academic results from students and fairest use in education. Therefore, investment would be reverted in better and larger human capital. The results of this particular chapter are a guide to improve the distribution of resources especially in education at departmental and municipal level 2 ; likewise it is a basis for designing policies to attract or retain people for municipalities and even departments. Also, it is an instrument to plan the future urban development of the cities and municipalities to ensure dwellings with adequate utilities. Therefore the objective of this chapter is to determine the migration patterns of Colombia; in order to accomplish it the following methodology is adopted: first, characterising migration at national, departmental and municipal levels; second, describing the migration trends inside of departments and across them; and finally establishing the features of the sending municipalities and the receiving municipalities. This part uses the model of the Office of the Deputy Prime Minister of United Kingdom (ODPM 2002) to analyse the spatial interaction in this country. The data used in this research is the record of the Ministry of Education enrolments from 2006 to 2009, which include all the students in official school (72% of students in Colombia). Although, the data is restricted to students between 5 and 17 years, which is a limitation of this study it offers a very clear picture of the internal migration in Colombia and the results coincide with the national trends described by other authors. In addition, this is the first time that the data of Ministry of Education is utilised in a socioeconomic study. 2 Departments are as states in the United States and municipalities are as towns. A set of municipalities conform a department. 2

3 In addition, this study is important because Colombia has at least 3.5 million displaced people, and at least 44% displaced men and 39.8% displaced women are between 5 and 19 years old (National Government, 2008). According to the Ministry of National Education (2001), displacement has caused serious problems including low educational level, emotional trauma, lack of documentation and certificates, child labor and emotional blocks worsening learning difficulties. The migrants, besides experiencing a change in their geographical location, face occupational, social and cultural changes. Additionally, if they are student, they may face admission barriers to the education system in the destination municipality such as: lack of places, change of academic schedules or teacher s availability, among others. Moreover, such students have to get used to environmental conditions prevailing in the new school, live with their new school-mates, and acquire knowledge and abilities efficiently. But if their migration was caused by enforced displacement, it is very probable that they face family disintegration, homes with a single parent, absence of attachment and communication, psycho-social dysfunctions and denial to the so-called displaced by the conflict (MinEducación 2001). Besides the aforementioned, many displaced people are peasants that due to exogenous reasons suddenly ran away without any previous plan, and they often face work and poverty in the city. Their accumulation of human capital is generally smaller than that of the poor residents of the destination municipality. According to the Ministry of Education, 35% of the displaced adult population have no education at all, while according to Sisbén, 14% of poverty-ridden adults lack any education. Moreover, three quarter of displaced people are below the indigence line 3 (Ibáñez, 2008). Four of the most important results are: 1) in Colombia, migration does not imply an improvement in life conditions, since most of students move to lower socioeconomic stratums regardless of the zone (urban or rural) of their origin or destination; 2) the effect of wealth at municipal level in Colombia agrees with the theory of the migration hump; 3) the 3 The poverty line represents a level of household income just adequate to meet the basic nutritional and nonfood needs of all its members. The extreme poverty line (or indigence) represents a level of household income that can not adequately cover the nutritional needs of all its members. Source: Juan Luis Londoño and Miguel Székely, "Persistent Poverty and Excess Inequality in Latin America, ," Working Paper 357, Inter- American Development Bank (IDB), Washington, DC October

4 effects of violence and unemployment on out-migration are conditioned to the wealth of the municipality and 4) migration flows from municipalities with armed conflict and/or illicit crops are addressed to municipalities with worse living conditions in most cases. Colombian territory is divided into six natural regions with geographic, economic and demographic different from each other. Related to its political division, Colombia is divided in 32 departments, and each department is divided in municipalities. Departments are like states in the United States and municipalities are like towns. The calculations presented in the article are computed for both political divisions. In short, a set of municipalities is a department and a set of departments is a region. In addition, each one of these divisions has rural and urban areas called zones. The urban zones are characterized by being composed of groups of buildings and structures grouped into contiguous blocks, which are bounded by streets and avenues mainly. The rural zones are characterized by the provision of scattered dwellings and farms, and they do not have streets, highways, avenues and others. Since the entire article makes reference to the departments and regions to explain the migration patterns, two maps de Colombia are provided in the Appendix 1 (departmental and regional division). 1. Internal migration in Colombia 1.1 Internal migration history and spatial distribution of the Colombian people The last century in Colombia has been characterised by large demographic changes. First, according to the Census in 2005 the Colombian population was 41.5 million people, which is ten times the population of 1905 and twice the population of Second, in the twentieth century, Colombia experienced virtually the entire process of demographic transition, from high birth and mortality rates to low rates (Flórez 1990); and third, Colombia became predominantly urban instead of rural, where the population is concentrated in the main cities. This last change is mostly caused by migration between zones (rural and urban) and among regions. The percentage of urban population has increased over the years; from 31% in 1938 to 69% in 1993 and 76% in This increase in urban population was primarily due to the net 4

5 transfer of rural population to urban locations (Chackiel and Villa, 1993 cited by Florez, 2000). This transfer was caused by migration and reclassification of localities. Rural - urban migration is selective by age and sex. According to Florez (2000), migrants have been mainly teenagers and young adults between 15 and 30 years old, mostly women, who seek better job opportunities in cities. As a result, there is an imbalance in gender distribution between areas (Flórez 2000). According to the Census 2005 in urban areas the percentage of men was 48% and the percentage of women was 52%, while in rural areas these were 53% and 47%, respectively. McGreevey (1968) indicates that the migration process in Colombia occurs in two stages. The rural migrants first go to regional urban areas and then migrate to large cities. Thus, between 1938 and 1973 the four largest cities (Bogotá, Medellín, Cali and Barranquilla) increased the percentage of total population living in them from 9% to 25%, forming an urban network, in which none was more important than the other (Florez, 2000). However, in the 1980s Bogotá became the principal economic centre of Colombia; as a consequence, in % of the Colombian population lived in Bogotá and 13% in the other three cities (Map 1). The earlier process was accompanied by the consolidation of metropolitan areas around large cities, a consolidation that still persists; for instance, Soacha and Bello 4 rank third and fourth (after Map 1. Population of Colombia Census Soacha and Bello are municipalities that are part of the metropolitan areas of Bogota and Medellin, respectively. 5

6 Bogotá and Medellín) in the ranking of destinations of students between 5 and 17 years old who migrated between 2006 and Currently, most of the population of Colombia is located in the northern, central and western parts of the country, with a municipal average of 198 inhabitants per square kilometre and a standard deviation of On the other hand, the Orinoco and Amazon regions with Choco (Map 1) are the areas with much lower population density with a municipal average of 11 inhabitants per square kilometre and a standard deviation of 17 (Census,2005). 1.2 Voluntary and forced migration in Colombia Cushing (1999) states that geographical mobility is generally classified as voluntary and it is closely related to opportunity; it is in fact considered as a means through which families try to improve their well-being. On the other hand, Ortiz de D'Arterio (2004) applies the Pushpull theory proposed by Everett Lee in 1966 when summarizing the reasons for emigrating: migrants react to economic expulsion forces such as unemployment or underemployment, hunger, low productivity, etc. or social factors such as lack of safety and services. Likewise, they respond to positive factors in destination areas; for instance, industrial and tertiary employment opportunities, higher salaries, better education opportunities, social services and recreation, among others. Compared to this model the Colombian case is very particular, since there is voluntary migration looking to improve well-being, as well as enforced migration to flee from violence. Voluntary migrants, whether poor or not, seek to improve their life conditions through access to services, education, social security, etc. The migratory flows of the poor on the one hand, although reflecting poverty in origin areas, also respond to certain economic incentives from a population actively looking for means to generate complementary income. These povertystricken families try to reduce labour supply excess and seek to maximize the use of the labour force. Molinas (1999) who studied internal migration in Paraguay, concludes that migrants main destinations are those areas where they can maximize their employment probabilities. Regarding enforced displacement, Colombia takes second place at the world level, after Sudan, with 7,8% (3.5 million people) of the country s population affected (Social Action, 2007). Violence originates from three armed organizations, currently branded as terrorists: 6

7 the Colombian Revolutionary Armed Forces (FARC), the National Liberation Army (ELN) and the Colombian United Self-defenses (AUC). Therefore, Colombia features not only voluntary mobility, but also enforced mobility or displacement, objective of which is to flee from violence. Displaced people are not migrants taking an economic option in order to look for better horizons in main towns, but families fleeing from violence threats exercised by armed groups to evacuate their territories: massacres, selective murders, rape, compulsory recruitment of children, kidnapping and anti-personnel mines (Salomón Kalmanovitz, 2008). The magnitude of the effect on geographical mobility can vary according to home features and the reason for migration. Voluntary migrants and displaced (involuntary migrants) are two different populations. Generally, poor voluntary migrants are young with low education and few people to care for (Molinas, 1999), while 90.6% of displaced migrate with the complete family nucleus and one-third involve an extended family (Ibáñez, 2008). Voluntary migrants go to areas where they have a higher probability to obtain work and increase their income (Molinas, 1999); in contrast, 65% of displaced households choose the municipality of reception because they have family and friends living there (Ibáñez, 2008). In addition, it is possible that the two stage process of McGreevey (1968) 5, which makes migration from countryside to city less traumatic, has been disrupted for the displaced victims of violence, whose principal destinations are the major cities; in fact, Ibáñez (2008) states that 78.3% of displaced households consider migration as definitive and prefer to migrate to the final destination directly. 2. Migration Measurement for Students of Colombia: national, urban and rural levels and considering socioeconomic conditions and armed conflict 2.1 Migration Measurement using the Ministry of National Education Data of Colombia (SIMAT) The data used in this research are records of school enrolment by the Ministry of Education from 2006 to The basic variable taken from the SIMAT data is the place of residence of the student, which may be different to the place where she studies. SIMAT data contain all the students of public schools in Colombia (72% of the total), who are 10 millions approximately and 97% of them are between 5 and 17 years old. Because of 5 The rural migrants first go to regional urban areas and then migrate to large cities. 7

8 this, this study selects just the students in this age range. The first step is to merge the data for consecutive years. The next table shows the number of students found in two consecutive years; however, there is a loss of information, the percentage of merging is 75% on average. Table 1 Results of the merging procedure Years # students found in both years Merging percentage 2006 y ,420, % 2007 y ,062, % 2008 y ,102, % Since this loss of information, a logit model is estimated for each period. The dependent variable is 0 if a student is not found in both years and 1 if he/she is found. The independent variables are taken from the first year of the period. The results of the models are in Appendix 2. According to them the merging process is not random; however, this could be caused by the number of observations, which is more than eight millions. Because of this, a weight for each student who merges is calculated using the probability estimated by the logit models. Then two municipal out migration rates (the dependent variables in the first empirical model) are estimated, one using weights and the other without them. The correlation between both out migration rates is 0.94, which means that they are practically the same variable. In addition, the models of Section 6 are estimated with both out migration rates and their coefficients are very similar, thus just the fifth model with weights is kept it in order to compare both results. Therefore, the analysis done in the study is based on unweighted data. Appendix 3 describes the calculation of migration rates at the national, departmental and municipal levels that are presented in this and next section. 2.2 National Migration Measurement Table 2 shows the migration probability over the four years. There are two measurements of this probability for the period , because during this period an incredible 82% of the students of Cesar move to Valledupar, capital of this department. Because of this, the migration probability including Cesar is 1% higher than the probability without it. Other municipalities that presented an unusual behaviour during the same period are those that belong to Antioquia and Caquetá, where the guerrilla has an active presence; because of this, 8

9 they are considered in the calculations and this is the reason why the migration in is 0.8% higher than in the other two periods. Table 2 Migration Probabilities for students with ages between 5 and 17 years old Gender Years # students in t-n and t Migration Probability Municipalities (%) Migration Probability Departments (%) Total * ** Males * ** Females * ** Source: calculations made by the author from the Ministry of National Education Data (SIMAT) *Including Cesar ** Students whose ages were between 5 and 14 years old in 2006 In general, the annual municipal migration probability of students is between 2.3% and 3.2% and the departmental migration is between 1% and 1.6%. Although the migration probability for men is slightly higher than the probability of women, the difference is not significant, which is expected because the population of study is students among 5 and 17 years old who depend on their parents mostly. The probability of migration for the complete period is calculated taking the students whose ages were between 5 and 14 year in This probability is around 6.8% for the municipal level and is 3% for the departmental level. The mobility of children whose ages ranges between 5 and 17 years old depends on the mobility of their parents. Therefore, younger children are more likely to have younger parents, who have not decided to settle in a permanent place; which explains the negative slope of the migration profile, especially at municipal level (Figure 1). The slope of the departmental profile is less pronounced because departmental migration is not as common as municipal migration. The municipal migration for the periods and overlaps, and the curve for is higher them because of the reason given previously (municipalities of Antioquia and Caquetá with armed conflict and high out-migration are 9

10 included). The municipal and departmental profiles have a different behaviour in , suggesting that departmental migrations are more common in the last period. Figure 1 Municipal Migration Probability Profile Figure 2 Departmental Migration Probability Profile 2.3 Migration Measurements according to type of zone: urban and rural The information of type of zone where a student lives is available for 2008 and 2009, so the analysis of this section is limited to this period. In this case, the municipal and departmental migration of men in urban areas is 12% and 7.7% greater than the migration of women, respectively. This could mean that urban male students are more prone to move and study away from their parents than females. In contrast, both migrations are practically equal for both genders in rural zones. The municipal mobility of rural students is 27% higher than the mobility of urban students; however, the departmental migrations are similar for the two zones. The departmental mobility of male urban students is higher than their rural congeners. Therefore, male students of urban areas migrate longer distance than those of rural areas. This is consistent with the theory of McGreevey (1968), which indicates that rural migrants first arrive to regional urban areas and then migrate to large cities. In contrast, female rural students migrate more and longer distances than female students in urban areas. 10

11 Table 3 Migration Probabilities for students by zone Migration Probabilities Zone (%) Urban Rural Municipal Total Males Females Departmental Total Males Females Historically, the most common internal migration is from countryside to city, creating urban areas in large settlements with very low quality of life; usually these settlements are around cities. Nonetheless, more than the 50% of migrant students move between urban areas. Table 4 shows how students move between zones. In general, migration to urban areas is 3.3 times the migration to rural areas. Moreover, 77% of the migrant students who live in an urban area in 2008 move to an urban area in 2009; in contrast, just 41% of rural migrant students in 2008 move to rural areas the next year. Related to this, the United Nations projections indicate that the number of city-dwellers will continue to increase from the 3200 million people today (almost half of world population) to nearly 5000 million in 2030 (United Nations 2007). Table 4 Migration between zones Year 2008 Origin Population Students who migrate Zone Urban Rural Number Urban Percentage Column Percentage Number Rural Percentage Column Percentage Total

12 2.4 Migration Measurement taking into account socioeconomic conditions Migration is often presented as an option for leaving poverty, so poor people are more prone to migrate than non-poor people. This section analyses this issue using socioeconomic stratum. Socioeconomic stratification involves the classification of residential properties into six groups through the physical characteristics of the immediate environment and urban or rural context in which they find themselves. Better conditions of a dwelling and of its environment implies a higher stratum. The strata are used to redistribute income on public services (water supply, energy, sewer, etc.) so the two higher stratums (the richest ones) subsidize the services of the three lower stratums. Table 5 shows that more than one- half of children in public schools live in stratum 1, and this proportion has increased over the years. In fact, the grow rate of students in stratum 1 for is 9.3% and for is 7.06%, whereas in it was 1.3%. On the other hand, the percentage of students in stratums 2 and 3 has decreased by 11.4% between 2006 and This could be a consequence of the economic crisis that the world faced in 2008 and In 2008 the Colombian economy grew at a rate of 2.5%, registering a strong decline of GDP growth compared with 2007, a year in which its growth rate was 7.5% (Mesa et al. 2009). On the other hand, the rural zone has an unexpected behaviour, since the proportion of students in stratum 1 diminishes between 2008 and Table 5 Percentage of Students by Socioeconomic Stratum Urban and Rural Urban Rural Stratum Municipal and departmental migration probabilities are shown in Table 6; these exhibit an inverse relationship with the stratum in the first two periods, but in the third period the relationship changes to being direct. This could also be attributable to the economic crisis. According to National Statistic Department and National Planning Department (2009), the percentage of people living in poverty diminished from 50.3% to 46% between 2005 and 2008, but the percentage of indigence increased from 15.7% to 17.8% and the most affected 12

13 is the rural zone, where indigence increased from 27.4% to 32.6%. As a result, in Colombia a non-negligible percentage of poor people become poorest in the last years, thus they cannot afford the cost of migration. In this sense Skeldon indicates: In apparent contradiction to the logic of survival migration, the general finding of most studies of migration in non-disaster situations is that it is not the poorest who move but those with access to some resources, no matter how meagre these might appear. Migration always involves some costs of transportation and the abandonment of many of the few possessions the poor might have. The poorest of the poor cannot afford either risk or movement and the majority starves in situ. (Skeldon 2003) However, in the two first periods students from stratum 1 and 2 migrate inter-municipally more frequently than students from stratum 3. Actually, in the municipal mobility of children from stratum 2 is greater than the mobility of children from stratum 3, but their departmental mobility presents a contrary behaviour. This means that the migration of the poorest students is likely to be mainly local or regional. However, for the last period in the rural area the municipal migration of stratum 1 is less than the migration of stratum 3 and their departmental migration are very similar. Table 6 Migration Probabilities by stratum of student between 5 and 17 years old Urban and Rural Urban Rural Stratum Municipalities Departmental In general, a higher percentage of students decrease their stratums than those who increase them, especially for the second period ( Table 7). The same result is obtained for the students who migrate, with the difference that in this case the percentages are much higher in comparison with all students. Observing Table 8, it is evident that migration does not imply an improvement in the life conditions, since in average 68% and 36% of students of stratum 3 and 2 respectively decrease their condition. In addition, the percentages of children who move to an inferior stratum are increasing over years. 13

14 Table 7 Stratum changes for all students (%) From 2006 to 2007 From 2007 to 2008 From 2008 to 2009 Stratum Stratum increases Stratum decreases Stratum increases Stratum decreases Stratum increases Stratum decreases Table 8 Stratum changes for students who migrated (%) From 2006 to 2007 From 2007 to 2008 From 2008 to 2009 Stratum Stratum Stratum Stratum Stratum Stratum increases decreases increases decreases increases Stratum decreases Table 9 shows the movements among stratums and zones. The results coincide with the conclusions of Table 7 and Table 8; nonetheless, this table reveals that if students move between urban and rural zones, most of the time there is a decrease in their stratum. For instance, 51.19% of the children in stratum 2 who move from urban to rural area decrease their stratum; the same phenomenon occurs with children in this stratum who move from rural to urban areas, of them decrease their stratum. This is an effect of the difference in the stratification between both zones. On the one hand, three quarters of the rural population is stratum 1, so it is very likely to change to a property with this stratum when a person move from urban to rural area. On the other hand, and this is just a hypothesis, the cost of living in stratum 2 in rural areas may be equivalent or similar to the cost of living in stratum 1 in urban areas. Finally, the students who migrated municipally, most of them reduce their stratum regardless of the type of zone of their origin or destination. 14

15 Zone 2008 Table 9 Stratum changes for students by zone between 2008 and 2009 (%) All students (migrants and no-migrants) Stratum 2008 Zone 2009 Students who migrated Stratum increases Stratum decreases Stratum increases Stratum decreases Urban Rural Urban Rural Urban Rural Urban Rural Urban Rural Migration Measurements taking into account the Armed Conflict The armed conflict variable that is used in this section is the attacks against the civilian population at municipal level, which include kidnapping, massacres, selective assassinations, forced recruitment and landmines. This variable is taken from the Human Rights Monitoring of the Vice Presidency of Colombia. According to Azam and Hoeffler (2002), the violence against civilians is motivated by military objectives. In Colombia, the guerrilla through this strategy has eliminated community leaders, terrorized civilians, expanded their territory, diminished the power of their enemies and usurped goods and properties(ibañez 2008). As a result, since 2002 on average 266,635 people per year have been displaced(ibañez 2008). Table 10 shows the migration probability in the municipalities that suffered attacks against the civilian population in the year indicated in the first column. When all the municipalities are taken into account, there is no difference between the probabilities of Table 2 and Table 10; even some of them are higher in the first table. This occurs because big municipalities, such as Bogotá and Medellín, are among those that suffered terrorist attacks. Because of this, the fourth and fifth columns present the probabilities for municipalities with population less than 100,000 and 20,000 inhabitants. In these cases, the migration probabilities are higher in a range from 40% to 185% than the probabilities in Table 2. In conclusion, the effect of the terrorist attacks on a municipality depends on the size of its population. 15

16 Table 10 Migration Probability in the municipalities that suffered attacks against civilian population Year of the violent action Period of Migration All affected municipalities Municipalities with population less than 100,000 Municipalities with population less than 20, SIMAT data include a variable called victim of armed conflict, which has four categories: displaced student, ex-guerrilla member, children of an ex-guerrilla member and she or he is not a victim. The distribution of students who are displaced between urban and rural areas is 66% and 34%, respectively; whereas, 72% of armed conflict victims are in urban areas and 28% in rural areas. Table 11 shows that three quarters of the students of the conflict victims and displaced live in stratum 1. Unfortunately the stratum of students before migration is unknown, so it is impossible to do the analysis made in the previous section. However, according to Skeldon (2003), in this case poverty is the result of the forced migration without adequate planning. For instance, the welfare losses of a Colombian household which is forced to migrate is 37% of the net present value of total household consumption during its life cycle (Ibañez and Velez 2007). In addition, according to Ibañez (2008) 94.4% of the displaced household are below the poverty line and 71.6% below of extreme poverty line. Table 11 Percentage of Students who are Displaced or Armed Conflict Victims by Socioeconomic Stratum post migration Displaced students Armed conflict victims Stratum

17 3. Geography of the migration of students in Colombia 3.1 Outmigration, Immigration and Net migration rates Most of the municipalities with the highest out-migration and immigration rates are located in the southern Meta, northern Caquetá, southern Guainía, northern Amazonas, northern Chocó, Putumayo, Guaviare and Vaupes (Map 2 and Map 3). They present outmigration and immigration rates higher than 4%. This phenomenon could be explained by the low population in the zone; for example, the departments of Amazonas, Guainía, Guaviare and Vaupes together represent just 0.55% of the Colombian population and Choco just 1.06% (Census, 2005). However, at departmental level Putumayo, Caquetá, Guaviare and Vichada also show the highest outmigration rate (Map 5) since they are framed in a complex context, the Amazon region. The departments of Amazon region 6 have exhibited a complex process of migration, since this is the site of recurrent clashes between the army and the guerrilla, as well as is the territory with the highest acreage of illicit crops. The sowing, harvesting, processing and marketing of coca are associated with a considerable migration of population attracted by the boom of this product or simply displaced from other areas of conflict (Gómez 1999). Besides, the people attracted by the boom are not rooted to the area, and with any signs of crisis they migrate to other regions searching better opportunities (Gómez 1999). In Map 6 the departments of Meta, Casanare and Arauca show immigration rates higher than 2%, and most of their municipalities have immigration rates bigger than 4% (Map 3), which is significant since the population of them is 1.3 million people. One factor that makes the migration process in the departments of Meta, Casanare, Putumayo and Arauca particular is that these departments have oil deposits, so they have a dynamic demography. For instance, Yopal (capital of Casanare) has a average annual rate of population growth of 9.5% between , 11% during and 6% in the period (Flórez 1999). Nonetheless, the effect of oil has decreased over time; the annual number of migrants to Casanare during is 1229 and in the period it is 621(DANE 2010) 7. 6 Putumayo, Caquetá, Guaviare, Vaupés, Guainía and Amazonas 7 National Department of Statistics 17

18 Outmigration, Immigration and Net migration rates municipal level Map 2 Outmigration rates Map 3 Immigration rate Map 4 Net migration rates 18

19 Outmigration, Immigration and Net migration rates departmental level Map 5 Outmigration rates Map 6 Immigration rate Map 7 Net migration rates 19

20 The zone called Uraba (Map 8), which corresponds to the northern part of Chocó and Antioquia, also presents high outmigration and immigration rates. This area has been characterised by its high displacement rate; some municipalities have lost more than the half of their population like Riosucio, 76% and Bojaya, 94.7% (Ibañez 2008).Comparing the results of the present study with the data reported by Social Action (Accion Social) 8, Map 2 with. Map 8, at least for this area both results are similar. Map 8 Reported displacement cases 2007 Source: SIPOD Accion Social Map 8 helps us understand the migration patterns of displaced people. Most of these people are below the line of extreme poverty, so they just flee to the closer municipalities, even if they have the same violence problem as the origin municipalities. Because of this, the dark zones of Map 2 are close or even coincide with the dark zones of Map 3, especially in the Amazon region and Chocó. In general, the net migration rate (Map 4) reveals a remarkable contrast between municipalities that belong to the same department. For instance in Chocó, the net migration rate of Riosucio is - 10% during , while Belen de Bajira presents a rate of 27%. Similar cases are found in Vaupes, Guainía, Bolivar, Putumayo, Meta, La Guajira, Caquetá, and Guaviare. Regarding the migration preference, departments such as Antioquia, La Guajira, Santander and Valle have the lowest outmigration rates (Map 5), which is in line with the percentage of students who migrate inside of these departments during According to the Origin- Destination Matrix the percentages of students who move from one municipality to another 8 The Presidential Agency for Social Action and International Cooperation is the entity created by the Government to channel the national and international resources to implement all the social programs that depend on the Presidency of the Republic and serve vulnerable populations affected by violence, poverty and drug trafficking ( 20

21 inside of Antioquia, La Guajira and Santander are in order of 79.93%, 61.22% and 61.43% (Appendix 4A). At departmental level (Map 6) Cundinamarca, Guaviare and Meta present the highest net immigration rates. The Origin-Destination Matrix, Appendix 4B, indicates that 55.6% of migrant students to Cundinamarca come from Bogotá. In the case of Meta, 27% come from Bogotá, 18% from Cundinamarca and 9% from Caquetá; and in the case of Guaviare, 21% come from Bogotá and 34% from Meta. In fact, the departmental average percentage of students who come from Bogotá is 14%. In addition, Meta shows high immigration rates because this department has regions with illicit crops, presence of guerrilla, oil and is close to Bogotá. In the rest of the country the municipal outmigration and immigration rates range between 0% to 4% (Map 2 and Map 3). The Amazon and Orinoco region and Chocó present the biggest loss of students; most of the departments of the northern, central and western of Colombia have negative net migration rates, with the exception of the richest departments (Cundinamarca, Bogotá, Antioquia, Valle and Atlántico) and of the department with oil resources. Also North of Santander has a positive net migration rate, which is caused by the increase of the student population in the municipalities next to the frontier with Venezuela. The high net migration of La Guajira is due to the immigration to Rioacha (capital) and Manaure, the latter being the place of a huge salt mine, whose production is one million tons a year approximately(aguilera 2004). At harvest time in Manaure salt mine, manual operation occupies about 2,000 indigenous Wayuu, between May and September each year; indirectly it benefits over 15,000 persons who provide transport services, maintenance and repairs of machinery and equipment, among others (Aguilera 2004). 3.2 Destination rate Historically, Colombia has four migration catchment areas, one for each major city: Bogotá, Medellín, Cali and Barranquilla (Flórez 2000). According to Flórez (2000), the migration catchment area of Barranquilla is exclusively caribbean and sparsely populated; the area of Cali is extensive and receives people from diverse departments; the area of Medellín receives people from Antioquia especially; and Bogotá has the most extensive area, which is the most densely populated. 21

22 Map 9 show the municipal destination rates per students; the rates of Bogotá and Medellín are larger than 4000 per students; then Soacha and Bello, which belong to the metropolitan area of the two first cities, have rates of 2043 and 1427, respectively. Subsequently, Cali and Villavicencio appear with the same rate of The list follows with the capitals of Risaralda, Bolivar, Santander, Norte de Santander, Huila, Tolima, and finally Atlántico. This means that Barranquilla (capital of Atlántico) has lost importance and a new city has taken its place, Villavicencio. Barranquilla is the largest industrial city and port in the Colombian Caribbean region; however, its catchment area has lost importance over time. Atlántico has seen a progressive loss of population; in the period this department lost 4700 people a year (DANE 2010). Its neighbours, Magdalena and Bolivar, during the same period lost and people a year, respectively (DANE 2010). According to the Origin-Destination Matrix the percentages of students who move from one municipality to another inside of Atlántico is 57%. The principal origins of the migrant students to Atlántico are Bolivar (23%), Magdalena (21%) and Cesar (9%). Cali is the capital of Valle, has a privileged geographical location due to its proximity to the port-city of Buenaventura, which is Colombia s main port on the Pacific Ocean, and is the third principal industrial centre of the country. However, 6500 persons left Valle annually between 2000 and 2005; that is the opposite of the trend for the period , in which 9500 people per year entered this department (DANE 2010). The student internal migration corresponds to 52%. The main origin department of the students who migrated to Valle between 2006 and 2009 are: Cauca (22%), Nariño (12%), Risaralda (10%), Antioquia (9%) and Bogotá (9%). Medellín has textile industry and is the second industrial centre in Colombia. During the period , 7770 people entered Antioquia annually (DANE 2010). The metropolitan area of Medellín consists of nine municipalities; one of them is Bello that ranks forth as destination. The student migration inside Antioquia is the highest across departments and corresponds to 79.93%. The principal origins of the students who immigrate to Antioquia are: Cordoba (12%), Valle (10%) and Chocó (9%). 22

23 Destination rates Map 9 Municipal destination rate Map 10 Departmental destination rate Map 11 Destination rate for displaced students 23

24 Bogotá is the main economic and industrial centre of Colombia. In the period , the commercial GDP in Bogotá represented 25% of GDP National Commercial (Planeación- Distrital 2007). In economic terms it is impossible to separate Bogotá and Cundinamarca, not only because Bogotá is the capital of this department but because of the dynamics between them. For the first half of 2008 in the Region Bogotá Cundinamarca 9853 new companies were created, with US$ 600 million in assets (DANE 2008). In addition, demographically one depends on another; during the period , the population of Bogotá increased by persons a year and the population of Cundinamarca by 6723(DANE 2010). The Origin- Destination Matrix indicates that 34% of the students who arrived in Bogotá in came from Cundinamarca, and 56% of the students who moved to Cundinamarca came from Bogotá. In this part, it is essential to highlight the role of Soacha 9, which is the principal receiver of people who come to the capital and ranks in the third position as destination after Bogotá and Medellín. Villavicencio is the capital of Meta and is called The Gate to the Llanos 10. It is an important centre of transit since most of the crops and livestock produced in the eastern region must pass through this town to get to the rest of the country. It is also the gate to the oil zone, as well as the coca-growing area. As Domínguez (1999) explains, referring to Putumayo and Meta: oil departments are, in greater or lesser dimension, coca producers. The population of Meta increased by 4300 inhabitants a year during (DANE 2010). In the period , Meta received students mainly from: Bogotá (27%), Cundinamarca (18%), Caquetá (9%) and Casanare (7%) (Appendix 4B). At the national level 52% of the students migrate inside of the same department. Table 12 presents the preference of destination by department (without taking account of the migration inside of the department). The table highlights that the migration patterns among departments are especially determined by distance, since most students migrate to closer departments with exception of Amazonas. It is worth clarifying that some preferences are stronger than others; for example, migrants from Tolima have clear preference to migrate to Bogota; but migrants from Antioquia do not have strong preferences (Appendix 4A). 9 Municipality that belongs to Cundinamarca and is next to Bogotá 10 Llanos is the vast savannas that lie between the Andes range and the Amazon rainforest 24

25 Table 12 Main Destination by Department Department Main destination % migrant students Department Main destination % migrant students Chocó Antioquia Bogotá Cundinamarca Cordoba Antioquia Caquetá Huila Valle Antioquia 8.75 Cesar Magdalena Bolivar Antioquia 8.28 Atlántico Magdalena 7.98 Cundinamarca Bogotá Vichada Meta Tolima Bogotá Guaviare Meta Boyacá Bogotá Guainía Meta Huila Bogotá Casanare Meta Meta Bogotá Vaupes Meta 6.37 Amazonas Bogotá 9.27 Putumayo Nariño Santander Bogotá 8.24 Arauca Norte de Santander 6.40 San Andres Bolivar Caldas Risaralda Sucre Bolivar N. Santander Santander 9.87 Risaralda Caldas Cauca Valle Magdalena Cesar Nariño Valle La Guajira Cesar Quindío Valle Antioquia Cordoba 2.55 Map 11 shows the destination rate for displaced students. The destinations for them are the traditional ones (Bogotá, Medellín, Cali and Villavicencio) plus some municipalities that are near to the conflict zones, especially capitals. These municipalities are: capitals of Caquetá (Florencia), Magdalena (Santa Marta), Cesar (Valledupar), Cordoba (Monteria), Putumayo (Mocoa), Sucre (Sincelejo), Tolima (Ibague), Chocó (Quibdo), municipalities of Northern Antioquia and Buenaventura. The problem is that municipalities have a limited budget to spend on their population, and armed conflict has added the displaced population which worsens life conditions and especially hinders access to social services like health and education. 4. Methodology Migration is the change of usual residence by a person or a set of persons during a specific time interval. This change implies two decisions, which do not have a chronological order necessarily: 1) migrants decide to leave an origin and 2) they choose where they wish to live. This means that the migration decision involves two stages. The former takes into account the factors that make the persons from a certain origin more likely to migrate to another region. On the other hand, the later stage considers the characteristics which make a destination more or less attractive. In others words and in the sense of migration patterns, the first stage investigates the characteristics of the sending municipalities aspect (push), and the second the receiving municipalities (pull). 25

26 The Office of the Deputy Prime Minister (ODPM 2002) presents a methodology to analyse the spatial interaction pattern in United Kingdom that contemplates the two stages. In the first stage, the authors estimate a model to predict migration outflows from each area; whereas, in the second stage, they determine a model to approximate the migration flows from origins to destinations. 4.1 First Stage In the ODPM research, the general model for the first stage is a model of the equation:, where is the out-migration rate, and and are the vectors of characteristics of origin i and the surrounding area of i, respectively. The attributes of the surrounding area of i ( ) is computed using a formula that gives more weight to the characteristics of the areas whose proximity to i is greater. In this sense, in order to obtain the variables Y it for every area, Fotheringham et al. (2004) applied the next formula: Equation 1 Where and are the variable X in the regions i and j, correspondingly; while, is the distance between both regions. The value of is in the interval ( ); however, the advisable value according to experience is -2. Values of values of Y across the regions; in contrast, a greater than -2 tend to produce homogenous <2 tends to take into account just the region closest to i, instead of calculating a local average of X in nearby location (Fotheringham et al. 2004). Thus, the weight of an observation j in Equation 1 decreases as the distance between i and j increases. The interpretation of Y depends on its value relative to unity: Table 13 Interpretation of neighbouring variables Value of Y Interpretation Y>1 The feature X is in general superior in neighbouring areas Y 1 The feature X is in general similar in neighbouring areas Y<1 The feature X is in genera inferior in neighbouring areas Source: Fotheringham et al. (2004) Regarding the model for estimating the first stage, three kinds of models are evaluated: OLS, Beta and the Quasi-likelihood Method (Papke and Wooldridge 1996). The three models provide 26

27 very similar predictions; however, the model that fits best is OLS since it presents the lowest sum of squared residuals. In short, the first stage is OLS models whose dependent variable is the out-migration rate between 2006 and The models are estimated with the municipalities whose population is above 2000 persons and with non-missing information on all the independent variables. In addition, models using clusters by departments are estimated, since municipal out-migration rates within departments are likely to be similar to one another. In order to see how much of the variability is within departments versus how much is between departments, an intraclass correlation is computed 11. This correlation is 0.45, which demonstrates there is a correlation inside of departments. 4.2 Second Stage In the second stage, a Spatial Interaction Model is estimated for each origin, thus each model expresses the preferences of the migrants of a specific region in choosing their destinations. The dependent variable is the number of migrants from origin i to destination j (Fotheringham and Wegener 2000; ODPM 2002): Where is the number of students who migrant from i to destination j, is the total number of migrants from region i, it attractive or unattractive, and represents one of the p characteristics of j, which makes is the distance between i and j. Since the model is estimated for each region i, the term is constant, therefore the final model is (ODPM 2002): 11 The coefficient of intraclass correlation is the measure of the relative homogeneity of the scores within the classes in relation with the total variation (Haggard 1958). This correlation is calculated using the command loneway of Stata. E.A. Haggard, Intraclass Correlation and the Analysis of Variance (New York: The Dryden Press, 1958). 27

28 The dependent variable in this case is count data, the number of students who migrate from i to destination j, with predominance of zeros, small values and discrete nature. Literature recommends using the Poisson regression model to study such data (Greene 2008). In fact, Flowerdew andaikin (1982) suggest that migration data are Poisson distributed; because of this there are several authors who have used this distribution to model the migration destination choice 12. However, the Poisson model has been criticised because of its assumption that the variance of the dependent variable y i equals its mean. Therefore, there are many extensions of the Poisson model that relax this assumption, but the most common is the Negative Binomial model (Greene 2008). This kind of model takes into account over-dispersion problems, where the variance is bigger than the mean; which the case of migration data of Colombian students. In the Negative Binomial model has a Poisson distribution with mean ; where is an unobservable random effect that has a gamma distribution with parameters, and represents the variance of. Thus the unconditional distribution of is a Negative Binomial distribution, with density: This Negative Binomial distribution has mean and variance. If is zero, the Poisson variance is obtained. If then the variance is larger than the mean. Therefore, the negative binomial model is over-dispersed relative to the Poisson. In brief, a negative binomial model is estimated for each municipality. Nonetheless, municipalities with few migrants and destinations are not included due the lack of fitness with the negative binomial model. Thus, just municipalities with a population above 2000 persons, with 100 migrants or more, and with at least 50 destinations are analysed in the second stage. As a result, 460 of the 1100 municipalities of Colombia are studied. 12 Robin Flowerdew and Murray Aitkin, 'A Method of Fitting the Gravity Model Based on the Poisson Distribution', Journal of Regional Science, 22 (1982), A. Stewart Fotheringham et al., 'The Development of a Migration Model for England and Wales: Overview and Modelling out-migration', Environment and Planning A, 36 (2004), Keiji Yano et al., 'A Comparation of Migration Behaviour in Japan and Britian Using Spatial Interaction Models', International Journal of Population Geography, 9 (2003),

29 5. Variables that make Colombian municipalities attractive or unattractive The main reasons identified by several authors (Jaramillo and Cuervo, 1987, Urrutia, 1990) 13 to promote migration in Colombia are: the regional gap in the provision of public services and in spending on education and health, the industrial development and foment of the construction sector, mechanization of agriculture, concentration of land ownership, wage differentials between regions, the distribution of the road of communication and violence. Nevertheless, there is not information for each one of these aspects, in consequence the independent variables that are included reflect direct or indirectly the causes of migration in Colombia. Variables Stage 1 Wealth Index: Voluntary migrants, whether poor or not, seek to improve their life conditions in terms of access to services, education, social security, etc. Therefore, migratory flows, especially of the poor, reflect poverty from emerging areas. On the other hand, the recruitment of young people by the guerrilla is easier in poor areas, reinforcing the reasons for emigration from these areas (Ibañez-Londoño 2008). In order to capture this aspect, an average wealth index is calculated for each municipality using the 2005 Census. The Wealth Index measures the socio-economic level in terms of assets and public services of households rather than income or consumption. The 2005 Census collected detailed information on housing characteristics (floor material, sewage, etc) and the availability of certain consumer durables that are directly related to the socio-economic development (fridge, washing machine, etc.). For each household is assigned a value that is generated through the methodology of principal component analysis, then a municipal average is calculated. Percentage of rural population: During the twentieth century Colombia experienced an urbanization process, especially in the 1950 s and 1960 s, and it was less pronounced in the last three decades (Flórez 2000). However, the armed conflict has motivated rural population to move to urban areas where they feel safe. Ibañez (2008) indicates that guerrilla has displaced 13 Samuel Jaramillo and Luis M. Cuervo, La Configuración Del Espacio Regional Colombiano. Tres Ensayos. (Serie Estudios 1; Bogotá: CEDE, Universidad de los Andes, 1987). Miguel Urrutia, 40 Años De Desarrollo Social. Su Impacto Social. (Bogotá: Banco de la Republica, 1990). 29

30 thousands of people occupying illegally their lands to clear the territory of potential opponents, expand territorial control and appropriate of valuable land. In consequence, there is a massive displacement of small landowners, with an abandonment of land between 1.7 million and 4 million hectares (World Food Programme 2001, cited by Ibañez 2008). This variable is computed at municipal level by the author using the Census Unemployment rate: Applying the Push-pull theory proposed by Everett Lee (1966), migrants react to economic expulsion forces such as unemployment; in particular, povertystricken families try to reduce family labour supply excess and seek to maximize the use of their work force. The unemployment rate is defined as the ratio between the people who are not working and are looking for a job and the economically active population; this population includes the persons who are working or are unemployed. The rate is computed using the 2005 Census. GDP per capita of 1998 (million of pesos 1975): GDP per capita reflects the economic activity of the municipalities, which is one of the principal characteristics that make attractive or unattractive a town. This information is provided by the Centre of Studies for the Economic Development of Los Andes University. Net attendance rate in primary education: It relates the number of children enrolled in primary education in the official age range with the population of this same range. 14. The educational conditions of the municipalities are essential since the objective population of this study is the students between 5 and 17 years old 15. In addition, a low attendance rates reflect lack of resources and economic problems of the municipalities (whether by government or civilian population), which is usually accompanied by low education quality. This rate is calculated by the author using the Census This is used as an indicator of universal access to primary education as a net rate equal to 100 percent means that all children of official age to attend primary school level are enrolled at this level. This indicator, unlike the gross primary enrolment rate, is not affected by situations of extra-age (enrolment in a grade above or below officially established for age) as this is given within the same educational level. 15 The Net attendance rate in secondary education is not included since its correlation with Net attendance rate in primary education is

31 Students studying in another municipality Percentage of students who study in another municipality: This variable reveals how students take advantage of the network of schools in their zone, which includes the schools of adjacent municipalities. This percentage also shows the quality of the transportation system in the region, since this factor facilitates the interaction between municipalities. Among the fifteen municipalities with most students studying in other municipalities, there are twelve that are neighbours of Bogotá, two municipalities bordering Medellín and Flandes, where many students study in Girardot. In contrast, among the fifteen municipalities with the least students studying in another municipality (0%), there are thirteen municipalities of Chocó (the poorest department). Moreover, the Figure 3 shows a clear direct relationship between the percentage of students who study in another municipality and the net attendance rate in primary school, which means that a high percentage of students who commute inter-municipally to study is not a symptom of lack of access to education in the municipality of residence. Figure 3 Students studying in another municipality vs. net attendance rate in primary education Net attendance rate in primary education Percentage of schools classified as high, superior and very superior by the ICFES exam in 2007: This variable indicates the quality of education in the municipality. The ICFES exam is the test that students take in the last year of high school and is used by the universities as the principal condition to select their new students. This variable is taken from the Colombian Institute for the Evaluation of Education (ICFES). Average Homicide Rate : The homicide rate, defines as homicides per persons, reflects the level of crime from a municipality, which is considered a push factor. The source is the National Police of Colombia. Total number of attacks between 2005 and 2007 against civilian population per 1000 inhabitants: In general, all studies confirm that the main reason of forced migration is direct attacks on civilians (Ibañez-Londoño 2008).The attacks against civilian population include massacres, selective assassinations, forced recruitment, kidnappings and terrorist actions. The 31

32 number of attacks is divided by the population since the impact of any terrorist act on the people is higher in small municipalities than in big cities. The information is taken from the Human Rights Monitoring of the Vice Presidency of Colombia. Distance to the capital of the department: Several researchers in Colombia have included this variable as one of determinants of guerrilla action 16. According to them, armed groups have more control over isolated populations than in economic and administrative centres, where the institutional presence of the State is poor. The distance used in the study corresponds to the Euclidian distance, since there is not information available about the real one. Altitude: In the international literature climate is considered as a push or pull factor. The Colombian climate is characteristic of the equatorial zone that maintains a uniform temperature for most of the year. However, the Andean mountain system crosses Colombia causing the greatest variety of climate determined by altitude. The climate levels are classified in warm weather (below 1,000 m, above 24 C, covers 80% of the country's area), moderate (between 1,000 and 2,000 m, temperatures between 17 and 24 C, corresponding to 10% of the country), cold (from 2,000 to 3,000 m, temperatures between 12 and 17 C, covering 8%) and heath land (over 3,000 m below temperature with 12 C)(IDEAM 2001) 17. Population density (people per Km 2 ): there are two reasons for including this variable. First, it indicates the size of municipalities; and second, low population density facilitates the actions of armed groups. Variables Stage 2 Stage two takes in all the variables used in stage 1 with two changes: a competing destination variable is added and population density is replaced by population in accordance with gravity models. Researchers such as Fotheringham et al. (2002), Roy (1990) 18 and Yano et al. (2003) 16 Ana Maria Ibañez-Londoño, El Desplazamiento Forzoso En Colombia: Un Camino Sin Retorno Hacia La Pobreza, ed. Ediciones Uniandes (Coleccion Cede 50 Años; Bogota, 2008), Fabio Sanchez and Mario Chacon, 'Conflicto, Estado Y Descentralización: Del Progreso Social a La Disputa Armada Por El Control Local ', Crisis States Programme, Working paper 70 (2005), Andrea Velásquez, 'The Formality in Property Rights: Determinant in the Military Strategy of Armed Actors?', Desarrollo y Sociedad, 61 (2008). 17 Institute of Hydrology, Meteorology and Environmental Studies of Colombia 18 The inclusion of the competition variable is one of the main advances in spatial interaction modelling (Jr Roy, 'Spatial Interaction Modeling: Some Interpretation and Challenges', Environment and Planning A, 22 (1990), ) 32

33 highlight the importance of including a variable that measures the competition that a destination faces from surrounding destinations. This variable is called the competing destination variable (Fotheringham 1986 cited in Yano et al. (2003)) and the usual way of calculating it is (Yano et al. 2003): Where k denotes a destination different from j, is the population of k and is the distance between j and k. Therefore, is one of the p variables that characterize each destination. Finally, Table 14 presents the descriptive statistics of the all variables, first and second stages. Table 14 Municipal statistics of the variables Variable Obs Mean Std. Dev. Median Min Max Out-migration rate Wealth Index GDP per capita 1998 (millions 1975) % of rural population Unemployment rate (%) Net Attendance Rate in Primary School (%) % of students studying in another municipality % Schools Classified as High, Superior and Very Superior Homicide rate (per 100,000 population) # Attacks per 1000 people Population density (people/km2) Distance to the capital of the department (km) Altitude (m) Cluster variable (competing destination variable) Population (million people) Results of the first stage: why do I leave this municipality? or where do migrants come from? The first stage seeks to explain why municipalities have different levels of out-migration. In terms of migration patterns, it answers the question: where do migrants come from? Thus, the dependent variable is the municipal out-migration rate. Appendix 5 shows the graphs between each independent variable and the out-migration rate In each graph is indicated the expected relationship according to the observed trend. There are two interesting results: first, the relationship between the unemployment rate and out-migration is negative; and second, the relationship with the wealth index depends on the wealth level. Trends of the rest of the variables are presented in the Table

34 Table 15. Observed tends according to the graphs Direct relationship with out-migration GDP per capita Percentage of rural population Homicide rate Number of attacks per 1000 inhabitants Distance to the capital of the department Inverse relationship with out-migration Net attendance rate in primary Percentage of students studying in another municipality % of schools classified as high, superior and very superior Altitude Population Density In Table 16, each model shows two standard errors: one corresponds to the errors without adjustment for clustering by department and the other with such adjustment. Model 1 includes all the variables explained in the previous section, without any nonlinear relationships or interactions. Model 2 has two additional variables: a dummy variable for the municipalities whose wealth index is higher than the median and the interaction between this dummy and the wealth index. The objective of this interaction is to estimate the two slopes that the graph of the wealth index versus out-migration rate shows (see Appendix 5). The objective of Model 3 is to find the thresholds where the relationships of unemployment rate and the percentage of students studying in another municipality with out-migration change their signs. The threshold in the case of unemployment rate indicates the point after which unemployment starts acting as a push factor. In the case of the percentage of students studying in another municipality, the threshold marks where the out-migration starts diminishing, especially among neighbouring towns, because of the high interaction between municipalities. The negative sign of the coefficient of unemployment rate is against to the theoretical push role of this variable; however, previous studies have obtained the same result. Some authors argue that using an aggregated measure of migration also aggregates sub groups whose motives for migration differ widely. Thus, the unemployed are summed with the employed and with individuals who are not part of the labour force; which underestimates or not reflect the incidence of unemployment (Greenwood 1997). In fact, the population of study are students and consequently their families, whose behaviour is less risky than young people, who venture to migrate first and then they look for a job. Besides, violence determines the behaviour of the displaced people since 90.6% of them migrate with the whole complete nucleus (Ibáñez, 2008). However, in the Model 1 the coefficient of unemployment rate is ; this means that for the 75 th and 95 th percentiles of this variable (6.14% and 13%) subtracts 0.38% and 0.82%, 34

35 respectively. In addition, Model 3 reveals that the threshold after which unemployment starts acting as a push factor is 29.4%, which is substantial. Regarding the wealth index, the wealthier municipalities (wealth index 43, this value is the median of the wealth index) the relationship between wealth index and out-migration rate is negative; in contrast, in the poorer municipality this relationship is positive. This result agrees with the migration hump, which indicates that at low levels of development (this means low income and wealth), there is low migration due the lack of resources; but as development rises, migration does too. Migration continues rising until a threshold level, after which migration begins to diminish since domestic economy offers opportunities to people (Stark and Taylor 1991 cited by Mendola 2006).(Mendola 2006) In the first model the relationship between wealth index and out-migration rate is positive; however, in the second and third models the slope for rich municipalities is negative. The slope for poor municipalities is ; in contrast, for wealthier municipalities is This means that the migration hump is not symmetric. The positive slope of poor municipalities is determined by the resources of the municipalities, the more resources the more out-migration. On the other hand, the negative slope for wealthier municipalities reflects declining outmigration as the quality of life in the municipality improves. Although GDP per capita has a positive relationship with the out-migration rate, its effect is small: if the GDP per capita increases by 1 million in 1975 values the out-migration rate increases by 0.009%, which equivalent to 0.3% of the average municipal emigration rate. In fact, for the 95th percentile of GDP per capita (45.98 million in 1975 values), this variable acts to increase the emigration rate by just 0.4%. 35

36 Dependent variable: Out-migration rate (%) Table 16 Results of the first stage Model 1 Model 2 Model 3 Coef. Standard Error Coef. Standard Error Coef. Standard Error OLS Cluster OLS Cluster OLS Cluster Wealth Index *** ** *** ** *** ** Dummy (Wealth Index>=43) *** ** *** ** Dummy (Wealth Index>=43)*Wealth *** * *** * Index GDP per capita 1998 (millions 1975) *** *** *** *** *** *** % of rural population *** ** *** ** *** *** Unemployment rate (%) *** *** *** *** *** *** Unemployment rate (%) *** *** Net Attendance Rate in Primary School (%) % of students studying in another municipality % of students studying in another municipality 2 % Schools Classified as High, Superior and Very Superior ** ** ** ** *** *** *** *** ** *** ** * * Homicide rate (per 100,000 population) *** *** *** *** *** *** # Attacks per 1000 people * * * * * Population density (people/km 2 ) * * Distance to the capital of the department (km) Altitude (m) *** * *** * *** ** Constant *** * ** * *** ** Number 913 R A A A A: adjusted R 2 36

37 If the percentage of rural population increases by 1%, the out-migration rate increases by 0.01%. At the 75 th and 95 th percentiles of the percentage of rural population (75.6% and 89.1%), this variable adds 0.87% and 1.02% to out-migration rate, respectively, which is not insignificant taking into account that the average municipal out-migration rate is 3.24%. Actually, the effect of the percentage of rural population could be even higher in the past when the most common internal migration was from countryside to city; however, according to Table 4, 65% of the migrant students come from urban zones; therefore, migration is not especially from rural to urban areas, but from urban to urban areas. The variable that measures the integration of a municipality with its neighbouring municipalities (% of students who study in another municipality) has a positive association with the out-migration rate, since more integration involves more means and links to migrate (for instance, the transportation system and jobs market). Indeed, this is the variable with the highest association with out-migration: an increase of 1% implies an increase of out-migration of 0.11%. This reflects the fact that most migration is over short distances and it depends on transportation system; even for some very close municipalities moving from one to another is almost equivalent to moving inside each one, because they behave as one municipality. Nevertheless, at the 95th percentile of the percentage of students who study in another municipality (6.16%), this variable adds just 0.62% to the out-migration rate. According to Model 3, the threshold where the out-migration starts diminishing by the high interaction between municipalities is 13.26%. An improvement of 1% in the attendance rate in primary education diminishes the out-migration rate by 0.013%. This effect is important since in the 25 th and 95 th percentiles of the attendance rate (84.97% and 96.55%) this variable diminishes out-migration by 1% and 1.13%, respectively. In contrast, an improvement in the quality of education has a marginal effect: an increase of 1% in the percentage of schools classified as high, superior and very superior decreases out-migration by 0.004%; even if 100% of the schools are classified in the best categories, its contribution to the outmigration rate would be just -0.04%. Clearly, the quality of the education system is not the principal aspect that makes out-migration increase or decrease across municipalities, even when the dependent variable is the emigration of students as in this study. In fact, Colombia still has serious problems of attendance especially in secondary (the municipal average attendance rate is 57%, Census 2005), so the quality of education is not the main priority. 37

38 Table 17 Results of the first stage Dependent variable: Out migration rate Model 4 Model 5 Model 5A: Out migration rate calculated with weights Coef. Standart Error Coef. Standart Error Coef. Standart Error OLS Cluster OLS Cluster OLS Cluster Wealth Index *** ** *** * *** ** Dummy (Wealth Index>=43) *** ** *** *** *** *** Dummy (Wealth Index>=43)*Wealth *** ** *** ** *** *** Index GDP per capita 1998 (millions 1975) *** *** *** *** *** *** % of rural population *** * *** * *** * Unemployment rate (%) *** ** *** *** *** ** Unemployment rate (%) 2 Dummy (Wealth Index>=43)*Unemployment rate Net Attendance Rate in Primary School (%) % of students studying in another municipality *** ** ** ** ** * *** * *** *** *** ** *** ** % of students studying in another ** * * * municipality 2 % Schools Classified as High, Superior and Very Superior Homicide rate (per 100,000 population) *** *** *** *** *** *** Dummy (Wealth Index>=43)*Homicide * * rate # Attacks per 1000 people * * ** *** ** *** Population density (people/km2) Distance to the capital of the department (km) Altitude (m) *** * *** ** *** * Constant ** ** R A A A

39 The violence variables, homicide rate and attacks against civilian population, increase out-migration. Although, the coefficient of homicide rate is small (0.0095), its contribution to the out-migration rate is important in the Colombia case, since 25% of the municipalities exhibit rates above 72 homicides per inhabitants. For instance, at the 90 th percentile of the homicide rate, this variable adds 1.1% to the out-migration rate. Regarding attacks against the civilian population, during 2005 and 2007, 25% of the municipalities of Colombia suffered attacks. The effect of this variable on the out-migration rate of these municipalities ranges between 0.006% and 2.47%. Finally, the coefficients of the last three variables (population density, distance to the capital and altitude) are not statistically significant or minor. An interesting fact found in this study is that the impacts of the unemployment and homicide rates are different for rich and poor municipalities. Because of this, both variables are interacted with the dummy for wealthy municipalities in Model 4 and 5. According to these, the negative effect of unemployment on out-migration in rich municipalities is almost the double of the effect in poor municipalities. Thus, it is possible that unemployment is more tolerable in wealthy municipalities than in poor. The case of the homicide rate is similar. The association between out-migration and this rate is positive for both kinds of municipalities, which is expected; however, its effect is higher in poor municipalities. The difference between Model 5 and 5A is the dependent variable; in the second regression, out migration rate is estimated by weighting students as Section 2.1 explains. However, as it is clarified in that section, from the two models it is possible to deduce the same conclusions. The last two models include the aspects of the surrounding area of each municipality: economy and violence. Just two variables are statistically significant: the wealth index and attacks against the civilian population per 1000 persons. Table 13 explains how to interpret these variables and in consequence their coefficients. The coefficient of wealth index of the surrounding area is positive; this implies that being surrounded by richer municipalities increases the out-migration of a municipality; which is logical taking into account that the distance of migration tends to be short. In contrast, the coefficient of attacks against civilian population is negative. This indicates that municipalities with fewer problems of violence than their neighbours have lower out-migration rates. 39

40 Dependet variable: Out-migration rate Table 18 Results of the first stage Coef. Model 6 Model 7 Standard Error Standard Error OLS Cluster Coef. OLS Cluster Wealth Index *** ** *** ** GDP per capita 1998 (millions 1975) *** *** *** *** % of rural population *** ** *** ** Unemployment rate (%) *** ** *** *** Net Attendance Rate in Primary School(%) % of students studying in another municipality *** *** ** *** % Schools Classified as High, Superior and Very Superior * Homicide rate (per 100,000 population) *** *** *** *** # Attacks per 1000 people ** ** ** *** Population density (people/km2) Distance to the capital of the department (km) Altitude (m) *** ** *** ** Wealth Index - neighbouring area ** * *** * GDP per capita 1998 (millions 1975) - neighbouring area % of rural population - neighbouring area Unemployment rate (%) - neighbouring area Homicide rate (per 100,000 population) - neighbouring area # Attacks per 1000 people - neighbouring area ** ** *** *** Constant Number R A A Results of second stage: What kind of municipality do I want to live in? Or where do migrants move to? In terms of migration patterns the second stage answers the question: according to municipality, to where do the people migrate? In order to do this, a negative binomial model is estimated for each municipality, 460 municipalities imply 460 models (for more details see section 4.2). If the coefficient of a variable X in the negative binomial model is , this means that a one-unit increase in X causes the expected count to decrease by a factor of exp( ) = , holding the rest of variables constant. 40

41 The summary of the results of the second stage are in the Appendix 6A. The first table of this appendix shows the results including all the variables explained in Section 5. The second table summarises the results excluding the variables that are not significant when the standard errors are adjusted by clustering by departments. The final results are presented in Table 19, in which just four variables describe the migration destination patterns. The first and second tables of Appendix 6A indicate that the destinations of some municipalities are municipalities with negative characteristics; for instance, with a high homicide rate; for 51.66% of the municipalities this is the case. This does not mean that dangerous municipalities are more attractive; it means that some of the most attractive municipalities, with better living conditions, have crime problems, such as Medellin and Cali. Medellin is the capital of Antioquia and the principal destination of the people of its department, so 77.3% of the municipalities where the homicide rate is significant (17 of 22) have the exponential of the coefficient more than unity. But the case of Cali (capital of Valle) is more extreme, just 1 of the 25 municipalities of Valle that participate in the second stage has the exponential of the coefficient less than unity. Similar results are obtained without taking into account the variables that are not statistically significant using clusters by departments. Therefore, the last model includes the four main determinants that are indicated by the two previous models: distance, percentage of rural population, population and the wealth index. Although, the wealth index has exponentials of coefficients that are both less and higher than unity, it is kept in the model since it summarises the public services conditions and assets of municipalities. Table 19 shows the results of the last model for all municipalities and Table 20 just for the municipalities that suffered attacks against civilian population between 2005 and Distance is the only variable that is significant for all municipalities. The coefficients of distance and percentage of rural population are always negative; in contrast, when population is significant, the sign of its coefficient is always positive. The coefficient of the wealth index is positive in 80% of the municipalities and in 62.22% of the municipalities with attacks in the period In fact, 34 of the 47 municipalities, in which the coefficient of the wealth index is negative, suffered attacks. 41

42 Table 19 Summary of the results of the second stage with the 4 principal determinants all municipalities Number municipalities with significant coefficients % municipalities with significant coefficients % Exp[coefficient] >1 % Exp[coefficient]<=1 Distance Percentage of rural population Population Wealth Index Total of municipalities 460 Table 20 Summary of the results of the second stage with the 4 principal determinants municipalities that suffered attacks against civilian population between 2005 and 2007 Models 3 Number municipalitie s with significant coefficients % municipalitie s with significant coefficient ts % Exp[coefficient] >1 % Exp[coefficient]<=1 Distance Percentage of rural population Population Wealth Index Total of municipalities 202 Map 12 shows the degree to which distance discourages the migration flows by municipality. If there is one origin and two destinations A and B, A is 1 kilometre farther from the sending municipality than B; the expected number of migrants to A is decreased by the factor that the map indicates, holding the rest of variables constant. For example, the expected number of migrants from Bogota to a destination 101 kilometres away is times less than the expected number of migrants to a municipality 100 kilometres away. The municipalities whose migration flows are less affected by distance specially belong to metropolitan areas; for instance, 15 of the 46 municipalities that are in the highest decile are capitals of a department, and the top 12 are 5 capitals and 7 municipalities belonging to metropolitan areas. This could be explained by the wealth of these areas, but also by their average education. The correlation between municipal average education and the coefficient of distance is 0.45, which means that migration flows from educated areas are less discouraged by distance. This coincides with the results of Levy and Wadycki (1974). Bogota and Leticia (Capital of Amazonas) have the longest migration flows; indeed more than the half of migrant students from Leticia move to Bogota. 42

43 At departmental level, in Antioquia, Boyacá, Risaralda and Santander the percentages of municipalities in the two first deciles of exp[distance Coefficient] are between 45% and 79%, which demonstrates that the migrants from these departments are reluctant to move long distances. Whereas, migrants from Atlántico, Bolivar, Cundinamarca, Huila, Nariño and Putumayo are prone to move long distances. The reasons for this behaviour could be idiosyncratic or geographical. For example, Antioquia and Santander have a high internal migration rate; 79% and 61%, respectively; this indicates that migration is strongly determined by their regionalism. In contrast, Boyacá has a strategic position; it is close to Cundinamarca and Bogota. In addition, Atlántico, Bolivar, Nariño and Putumayo are on the borders of the country, which explains their high coefficients. The eight municipalities that represent Amazonas, Guaviare, Guainía and Vaupes also have high coefficients because of their geographic position. The municipalities of the rest of departments are distributed across the five quintiles of the coefficient of distance. Map 12 The deterrence of distance 43

44 The geographic distribution of poverty in Colombia depends on the regional division, which is clearly illustrated by Figure 4. The Andean region is the richest and just 25% of its municipalities are below the median of the wealth index. Additionally, more than 80% of the municipalities of the Caribbean and Amazon region are in the two first quartiles of the wealth index. In the case of Orinoco and Pacific region, more than 60% of the Figure 4 Quartiles of the Wealth Index at municipal level by Region municipalities have the same condition; nonetheless, the behaviour of the Pacific region is improved by Valle, which is evident in Map 13. This map presents the wealth index at municipal level. The map highlights the strong poverty situation in the Pacific region (Chocó, Cauca and Nariño) and also in some municipalities of the Atlantic coast, especially in Sucre and Bolivar. Map 13 shows the Colombian economic geography. There are three foci: Bogota, Medellin and Cali. The economy of the country revolves around these cities as well as its wealth. The municipalities close to these cities also have high levels of wealth; but as the municipalities are farther from them their wealth decreases. At departmental level, Antioquia is a perfect example of the distribution of poverty: the darkness part of the department corresponds to Medellin and its metropolitan area, and the colour of the municipalities becomes clearer the further away they are. The same pattern is observed in most of departments such as Valle, Cundinamarca, Meta, Nariño, Cauca, Cordoba and Atlántico. Thus, three cities are at the centre of national economy (wealth), especially Bogota, and the capitals are the centre of the departmental economy. 44

45 Map 13 Wealth Index Map 14 Wealth Index Factors all municipalities Map 15 Wealth Index Factors- level of significance: 10% 45

46 Map 14 and Map 15 report the exponential of the coefficient of wealth index, and Map 15 just shows the coefficients that have 10% level of statistical significance. Most of the municipalities have exponentials of the coefficients of wealth index in excess of unity, 187 of 231 (Table 19), which means that inter-municipal migration flows generally are to municipalities with better economic conditions. Nonetheless, Map 15 indicates that there are several municipalities whose migration flows are to municipalities with poorer conditions (exp[coefficient]<=1), 47 exactly. These municipalities are located mainly in the Amazon, Orinoco, Caribbean and Pacific regions. Therefore, taking into account the poverty situation of these municipalities, it is likely that most of their population is below the poverty line, and so they can just migrate to closer municipalities, which are poor too. Besides, most of them are the centre of the armed conflict and/or production of coca, which worsens their situation further. Map 16 Municipalities with less than people that suffered attacks between 2005 and 2007 Map 17 Changes in the areas of coca cultivation in Colombia, Source: National Narcotics Directorate,

47 Map 16 and 17 show the municipalities that have suffered attacks and the areas of coca cultivation. Comparing the areas indicated by these maps with Map 15, the association between armed conflict and illegal crops with the migration patterns in the clearer zones of Map 15 is clear. In fact, these zones coincide more with the areas with illegal crops than with the zones with armed conflict for example the areas in Putumayo, Caquetá, Southern Meta, Vichada, Guaviare, Nariño and Arauca. Flórez (2000) states that during , the flows with Casanare, Guaviare and Putumayo as their destinations came mainly from Meta and the Pacific zone and were probably associated with the oil of Casanare and illicit crops in Guaviare and Putumayo. In contrast, in some areas that only have armed conflict, the exponentials of the coefficients of wealth index are not significant or are more than unity, such as Tolima and Huila. Map 18 presents the percentage of rural population. In addition to big cities, municipalities with oil have low proportions of rural population. At the regional level, the municipalities with high percentage of rural population (more than 70%) belong to the Amazon and Pacific regions, principally. At departmental level, more than the 60% of the municipalities of Cundinamarca, Boyacá, Santander and Norte de Santander are in the last two quartiles of this variable. Bogota, Atlántico and Quindío have the highest degree of urbanization because of their restricted area. Map 19 and Map 20 report the exponential of the coefficient of the proportion of rural population, and Map 20 just shows the coefficients that have attained the 10% significance level. The way of interpreting the information of the map is: the expected number of migrants from Cartagena to a destination with 50% of rural population is times the expected number of migrants from Cartagena to a municipality with 49% of rural population 19. This variable is significant in 90% of municipalities with a value always less than unity. Thus the percentage of rural population affects migration more from the clearer zones of Map 20. For instance, the effect of this variable is more notable in most of the municipalities of Amazon and Orinoco. Map 21 shows that the most populated region is the Andean region followed by the Caribbean and the Pacific regions. On the other hand, the Orinoco and Amazon regions with Chocó are the areas with the lowest population density, with a municipal average of 11 inhabitants per square kilometre and standard deviation of 17 (Census,2005). Map 23 presents the results of the binomial model and is interpreted as follows, for example: the expected number of migrants from Itagui to a destination with 2 million people is 7 times the expected number of migrants from Itagui to a municipality with 19 The expected number of migrants from Cartagena to a destination with 50% of rural population= * the expected number of migrants from Cartagena to a municipality with 49% of rural population. 47

48 1 million people 20. In the top 12 of the coefficients of population are 5 capitals: Popayan (exp(population)=933), Bogota (734), Cali (664), Monteria (500) and Bucaramanga (217), which means that in these capitals there are strong preferences to move to urban areas. Even though it seems that there is a pronounced direct relationship of the coefficient with their population, their correlation is just Actually, other capitals present low coefficients, such as Cucuta (8), Villavicencio (11), Manizales (16) and Santa Marta (27). Finally, there is not a pattern inside of the regions and departments. 20 The expected number of migrants from Itagui to a destination with 2 million people=7* the expected number of migrants from Itagui to a destination with 1 million people 48

49 Map 18 Percentage of rural population Map 19 Percentage of rural population factors all municipalities Map 20 Percentage of rural population factors - level of significance: 10% 49

50 Map 21 Population 2005 Map 22 Population factors all municipalities (Population in million people) Map 23 Population factors - level of significance: 10% (Population in million people) 50

51 8. Conclusion The migratory process of Colombian students offers a very clear picture of the internal migration in this country, in spite of this population representing a quarter of the total country. Because of the age of these students, 5 to 17 years old, their migration implicitly represents the migration of households with children in this age range. Additionally, some results obtained in this study coincide with conclusions of previous authors; in the same way, this research reveals new results of migratory flows at municipal, departmental and regional levels regarding pull and push factors such as distance, violence, rurality and wealth. In summary the most relevant results of this research are: In Colombia, migration does not imply an improvement in the life conditions since in average 42.2% students from stratum 2 and 3 move to lower socioeconomic stratums and 10% to superior stratum regardless of the zone of their origin or destination during The most preoccupying is that this behaviour has increased over years, which could be caused by the growing of indigence in the last three years. The effect of poverty in Colombia agrees with the theory of the migration hump: At municipal level, the poorest municipalities present low levels of migration; but as wealth rises, migration does too. This happens until a threshold level, in which the relationship starts being inverse because wealthy municipalities offer opportunities to people. At individual level, the poorest of the poor cannot afford the risk and cost of migration; but when their resources increase, their probabilities of migrating increase too, which is evident in the period , especially. Even though the most common migration is from and to urban areas, since Colombia is 76% urban, rural students have higher migration probabilities than urban students. Additionally, rural migration is characterised by short distance migration, while urban students and their families are more likely to move long distance. The same contrast exists between poor and non poor students: people from wealthy municipality tend to travel longer distances. The Amazon and Orinoco region and Chocó present the biggest loss of students and most of the departments of northern, central and western Colombia have negative net migration rates, with the exception of the richest departments (Cundinamarca, Bogotá, Antioquia, Valle and Atlántico) and departments with oil or salt resources. The principal destinations of students are Bogota, Medellin and Cali with their metropolitan areas, as well as Villavicencio, which displaced Barranquilla. 51

52 Most displaced persons are below the line of extreme poverty, so they just flee to closer municipalities, even though they have the same violence problems or worse conditions of living than origin municipalities. Therefore, in addition to the typical main destinations (Bogota, Medellin, Cali and Villavicencio), displaced prefer to migrate to the nearest capital. Regarding the results of out-migration models (first stage), an improvement of education in terms of attendance rates diminishes out-migration, but an improvement in its quality does not have an effect. The percentage of students studying in another municipality has the highest coefficient; however, its values are concentrated between 0 and 3%, so its effect is not relevant. In spite of Colombia being 24% rural, the percentage of rural population is associated positively with the out-migration and its incidence in 25% of municipalities is around 1%. Although the GDP per capita has a positive relationship with out-migration rate, its effect is small. Equally, the coefficients of population density, distance to the capital and altitude are not significant or minor. In regard to the variables of violence, they have the expected association with out-migration: more crime and attacks increase it. Their effects are important since the violence levels in Colombia particularly in small municipalities; in some cases these variables raise the out-migration rate by more than 1 percentage points. On the other hand, Colombia has zones where the armed conflict is conjugated with coca production and/or oil deposits, which act as pull factors, so poor people face a duality between fleeing from violence and participating in two of the most profitable businesses in the world. Because of this, in models of destination preferences (second stage) some municipalities have migration flows to zone with worse conditions; although in general inter-municipal flows are to municipalities with better characteristics (higher wealth index). The effects of unemployment and homicide rates on out-migration rates are mitigated by the wealth of the municipalities. This means that good standards of living compensate the low conditions in employment and crime in wealthy municipalities. The context of a municipality plays an important role too, because being surrounded by richer municipalities increases the out-migration rate of a municipality. On the other hand, municipalities with less violence problems than their neighbours present lower outmigration rates. This agrees with the fact that the distance of migration tends to be short. 52

53 The patterns of destinations by departments are defined by idiosyncratic and geographical reasons. There are departments, as Antioquia and Santander, where regionalism and within departmental migration are marked. Because of this, they present short distances migration. In contrast, municipalities of departments located on the borders have longer distance migration since their geographic position. Finally, the migration flows of the 460 municipalities included in the second stage are directed towards to more crowded and less rural municipalities. 53

54 Appendix 1 Map of the departments of Colombia 54

55 Appendix 2 Logit models for the merging process Estimate Odds Ratio Estimates Estimate Odds Ratio Estimate Odds Ratio Estimates Estimates Intercept *** *** *** Stratum *** ** *** Stratum *** *** *** Woman *** *** *** Displaced *** *** *** Morning *** *** Grade *** *** *** Rural *** Antioquia *** *** *** Arauca *** *** *** Atlántico *** *** *** Bolivar *** *** *** Boyacá *** *** *** Caldas *** *** *** Caquetá *** *** *** Casanare *** *** *** Cauca *** *** Cesar *** *** *** Chocó *** *** *** Cordoba *** *** *** Cundinamarca *** *** *** Huila *** *** *** Guajira *** *** *** Magdalena *** *** *** Meta *** *** *** Nariño *** *** *** N. Santander *** *** *** Putumayo *** *** *** Risaralda *** *** *** San Andres *** *** *** Santander *** *** *** Sucre *** *** *** Tolima *** *** *** Valle *** *** *** Quindío *** *** *** Amazonas and Orinoco Dependent variable *** *** ***

56 Appendix 3 The calculation of migration rates at the national, departmental and municipal levels A.1 Migration Probability Crude Migration Probability is defined as the ratio between the number of transitions or migrants and the population at the beginning of the time interval. For the Ministry of National Education Data (SIMAT) this ratio is calculated using the persons who were found in two points of time, thus it can be approximated as the number of migrants between t-n and t divided by the total of persons who were in both moments. T: count of transitions/migrants PAR: estimated population at the start of the time interval over which migration is measured A.2 Probability of migration for age This probability conserves the same logic that the Migration Probability, but it is calculated for agegroups and/or gender. In based on this, it is possible to graph age profile of migration. A.3 Migration Measurements by municipalities The calculations explain in this part are computed for the 1100 municipalities of Colombia, and the populations of almost the half of them are less inhabitants, so they are very sensitive to idiosyncratic migrations and to the low coincidence in the merging of the data between consecutive years of the SIMAT data Consequently, in order to obtain robust and consistent migration flows among municipalities the information for the three periods is aggregated. The next part explains how the migration rates are calculated using the aggregated information. a. Outmigration rate The out-migration rate is the ratio between the number of migrants from i and the average between the beginning and the final population of this region in the period. Using the SIMAT data and aggregating the three periods the populations are defined as: 56

57 It is important to notice that the initial population of does not coincident with the final population of , since the merging procedures for these two periods find different students (see Section 2.1). The migrants from i are the sum of students who moved to another municipality during the tree periods. b. Immigration rate The immigration rate is the ratio between the number of immigrants to i and the average population of this region between t-1 and t. The number of immigrants to i are the sum of students who arrived to i during the tree periods. a. Net migration rate The Net Migration Rate is the difference between the Immigration rate and Outmigration rate. 57

58 b. Destination rate The destination rate indicates the percentage of migrants who chose a specific region as a final destination. The number of migrants to i are the sum of students who arrived to i during the tree periods. 58

59 Appendix 4 A. Origin-Destination Matrix/ Row Percentage (to which department migrants move) 59

60 B. Column Percentage (from which department migrants come) 60

61 Appendix 5. Independent variables vs. Outmigration rate Wealth Index GDP per capita (million pesos 1975) Unemployment rate 2005 Percentage of rural population Net attendance rate in primary % Students studying in another municipality 61

62 % Schools classified as high, superior and very superior by the ICFES exam in 2007 Average Homicide Rate Total number of attacks between 2005 and 2007 against civilian population per 1000 inhabitants Distance to the capital of the department Altitude Population Density 62