Have Migrant Families Achieved Their Goals? Micro-level study on Migration, Education and Income in Latin America and the Caribbean Countries

Have Migrant Families Achieved Their Goals? Micro-level study on Migration, Education and Income in Latin America and the Caribbean Countries Namsuk Kim* May 2006 For a presentation at XIV International Economic History Congress, Helsinki, August 2006 Session 34. Migration, family and economy in industrializing and urbanizing communities Abstract Many countries in Latin America and the Caribbean (LAC) region are being urbanized and industrialized at a very high speed. The share of urban population grew from 65% in 1980 to 76% in 2001 and the share of employees in non-agricultural industry is now 28% (World Bank, 2006). Migrants from rural to urban area aim to get higher income for the family and/or better education for their children. Have they achieved their goal? The empirical result from 7 countries in LAC shows that younger generations of migrant families do have more educational opportunity, controlling geographic and individual heterogeneity. Migrant families are found to have higher income per capita comparing the non-migrant families. *World Bank, 1818 H St. NW, Washington DC, 20433, nkim@worldbank.org

1. Introduction Fast urbanization has been observed around the world in the last four decade. One third (32%) of global population lived in the urban area in 1960, and now almost a half (48%) of the population live in urban area in 2004. 1 As Figure 1 shows, the trend of urbanization is common across all regions. Among others, Latin America and the Caribbean (LAC) region is the highest urbanized area and the share of urban population in LAC is outstanding. Since fast urbanization reshapes the economy, it has been studied from various perspectives in the literature. Figure 1 Urban Population (% of Total Population) 90 80 70 60 50 40 30 20 10 0 1960 1990 2004 East Asia Europe LAC Middle East South Asia Africa World Source: World Bank (2006) The main incentive for migration in the literature has been argued to be a betterpaying job and better education opportunity. In 2004 Honduras, 56% of migration was due to employment related issues, and 29% of migrants moved to urban area to have higher education for themselves or their children. In 2001 Nicaragua, 31% of migration is due to employment and economic matters, while 32% is related to family decision regarding children's education. 2 Migrants move to urban area to achieve higher education and better employment despite the fact that they may have a risk to go through worse living standard of working poor in the big cities. But it seems that the risk of being worse off by migration can be overwhelmed by the aggregate positive effect of industrialization and urbanization of the economy. In order to study the benefit and cost of migration, the outcomes of migration must be carefully assessed with micro level household data, using the detailed information on migrant families. 1 World Bank (2006) 2 Calculated from the household level surveys. See the text for the sources. 2

This paper focuses on seven Latin America and the Caribbean countries and examines the hypothesis that migrants to urban area achieved higher education and higher income. The main contribution of this paper is to have a cross country analysis of the effect of migration on employment and income, controlling characteristics of migrants including the origin of migrantion. The origin town of migrants are identified as detailed as possible using the migration information in the data. The outline of this paper is as follows: Section 2 reviews previous empirical studies on the issue of migration. Section 3 discusses the data and estimation specification. Section 4 explains the empirical result and Section 5 concludes. 2. Literature Review A useful starting point for empirical studies of migration is the human capital model of migration (Mohlo 1986). This model takes migration as an "investment increasing the productivity of human resources, an investment [that] has costs and [that] also renders returns". Sandell (1977) and Mincer (1978) extend the model to the context of nuclear family with two wage-earners, arguing that net family benefit motivates the migration of a family, rather than net personal benefit. As the labor flow across countries grows, international migration is considered simultaneously with internal migration as in Stark and Taylor(1991). The findings from Mexico reported in the paper provide evidence that, if absolute income is controlled for, relatively deprived households are more likely to engage in international migration than are households more favourably situated in their village's income distribution. In contrast, the findings suggest an interesting 'income neutrality' result, unique to relative deprivation theory, in the case of internal migration. Recent studies focus on the potential motive for migration. de Jong(2000) argues that expectations the process of evaluating the chances for future attainment of valued goals in the home community (stay decision) vs alternative locations (move decision) along with family norms about migration are major predictors of intention to move, which in turn is a proximate determinant of migration behavior. Using longitudinal data from the 1992 and 1994 waves of the Thailand National Migration Survey, logistic regression models show that a strikingly different set of expectations, household demographic indicators, and migrant capital factors were significant determinants of migration intentions for men and women. Agesa(2001) uses data from Kenya to test the hypothesis that the internal migration may be a self-selection mechanism that sorts migratory and nonmigratory workers according to their characteristics. The findings lend strong support to the argument that migratory and nonmigratory workers may have sorted themselves accordingly to their characteristics. In particular, the results suggest that the incidence of migration is relatively higher for workers with a positive urban to rural earnings difference, suggesting that skilled workers self-select to migrate to urban areas. Other studies concentrate on the result of family migration. For instance, Cooke and Bailey(1996) finds that family migration increases the probability of employment among married women by 9 percent but has no effect on the probability of employment 3

among married men, from the empirical analysis focusing on a sample of married parents in the economically active population residing in the Midwestern US in 1980. This paper fits into the line of research on the effect of migration, not the cause of migration. This paper extends previous single country studies and conducts a cross country analysis of migration controlling the regional and personal characteristics of migrants. In the remainder of this paper, I focus on assessing the effect of family migration on three variables, the probability of completing primary or higher education, the probability of completing secondary or higher education, and the per capita monthly income of the household. 3. Data and Model Specification The data for this analysis are household level surveys in seven Latin America and Caribbean (LAC) countries: Dominican Republic, Guyana, Haiti, Honduras, Jamaica, Paraguay, and Nicaragua. 3 The seven countries may not be representative for LAC in terms of geographic location or the stage of economic development, but they share a common aspect of steady urbanization and disruptive growth. The uniqueness of the data in this paper is that the data have information on the migrants not only whether they migrated, but also the origin of migration, that is, from where they migrated. When we can identify a narrowly defined geographic area where the migrants moved from, the original conditions of migrants and nonmigrants are more likely to be homogeneous. Table 1 describes the household level data that are used in the analysis. My sample is composed of national censuses or surveys circa 2001 and the data includes the information on the residence of 5 years before the survey was conducted. The information on household income is available except in Guyana and Jamaica. Migration rates are calculated by the origin towns, and the average and standard deviation of the migration rates are presented in Table 1. 4 Table 1 Migration Data Year Type Number of origin towns of migration Dominican Income Data Average percentage of migrant population Standard deviation 2002 Census 32 3.28 1.64 Republic Guyana 2002 Census 10 N.A. 5.19 3.75 Haiti 2001 Survey 9 1.27 1.07 Honduras 2004 Survey 18 4.61 3.03 Jamaica 2001 Census 14 N.A. 0.11 0.05 Nicaragua 2001 Survey 126 3.82 3.38 Paraguay 2004 Survey 16 10.22 2.15 Source: Author's calculation from the data. 3 There are sixteen LAC countries that have information on migration in their micro level surveys, but only seven countries have detailed information on the origin and the timing of migration. 4 Detailed description is in the Appendix which is available from the author. 4

1) The source of Dominican Republic data is the national census, VIII Censo Nacional de Poblacion y Vivienda (CNPV) that was conducted in 2002. In the census, respondents were asked to answer in which providence they resided in 1996 and whether they were in urban or rural area. The origin towns are listed by 32 providences. 2) For Guyana, I use Census 2002 data that includes information about the residence of 5 years before the census. Ten regions and urban/rural information are collected in the census. 3) For Haiti, Enquete sure les Conditions de Vie en Haiti (ECVH) 2001 is used for the analysis. The origin residence is disaggregated only by 9 departments. Given the size of Haiti, however, department is not too broad geographic area for the analysis in this paper. Urban/rural area information is also available. 4) The data source for Jamaica is 2001 Census. The name of parish where they lived in 1996 is collected in the census. However, income data are not available. 5) Encuesta Nacional de Condiciones de Vida (ENCOVI) 2004 is used for Honduras. It is a national survey including the residence in 1999 and household income data. 6) I use Encuesta Nacional de Hogares sobre Medicion de Niveles de Vida (EMNV) 2001 for Nicaragua. It collects the name of prior residence where migrants lived in 1998 and the residences are disaggregated by 126 municipals. 7) For Paraguay, I use Encuesta Permanente de Hogares (EPH) 2004. 16 Department and Urban/Rural area can be identified for 2004, 1999, and the place of birth. Table 2 shows selective statistics of countries that are analyzed in this paper. The percentage of urban population suggests that Guyana, Haiti and Honduras are less urbanized countries comparing to Dominican Republic, Jamaica, Nicaragua and Paraguay. Many of theses countries have higher urbanization by the standard of Low and Middle Income Counties (43%) but not so much comparing to LAC average (77%), Europe average (76%) or United States (80%). 5 However, the speed of urbanization is remarkable. In four countries the average annual growth of urban population exceeds the average growth in Middle Income Countries (27%). Guyana and Jamaica both have slow growth of urban population but the reasons are different. Guyana doesn't have either high share or fast growth of urban population because the industrialization is at a low speed and the economy heavily depends on agricultural exports lead by sugar sector. But Jamaica is already one of the highly urbanized/industrialized countries in the Caribbean region and the geographical distribution within the country has been stable over the last decade, while international migration is gaining more significance. 5 World Bank (2006) 5

Table 2 Selective statistics of countries in LAC Country Dominican Republic Guyana Haiti Honduras Jamaica Nicaragua Paraguay Year 2002 2002 2001 2004 2001 2001 2004 Urban population 59.0 37.2 36.2 46.0 52.1 56.5 57.9 (% of total) Urban population 2.1 1.4 3.2 3.1 0.6 2.7 3.5 growth (annual %) GDP per capita 2510.3 984.6 454.2 964.6 3128.6 801.1 1372.5 (constant 2000 US$) Industry, 26.1 28.9 17.0 31.2 32.9 31.4 24.0 value added (% of GDP) Trade 84.5 198.3 48.5 90.8 92.6 70.9 73.2 (% of GDP) School enrollment, 98.7 99.2 75.5 87.5 90.1 82.8 89.3 primary (% net) School enrollment, secondary (% net) 40.3 29.1 31.2 30.2 76.0 36.6 51.1 Note: Industry covers mining, manufacturing, construction, electricity, water and gas. Source: Author's calculation from the surveys and World Bank (2006) Countries analyzed in this paper are rather low income countries, with GDP per capita less than $1000 in Guyana, Haiti, Honduras and Nicaragua. But the share of industry sector in the total GDP reaches up to 32%, which is higher than world average (28%). Only Dominican Republic and Jamaica have lower share of industry sector comparing to 1990, while the share of industry increased in the other countries. Therefore, Guyana, Haiti, Honduras, Nicaragua and Paraguay are at the early stage of industrialization and on the path of expanding industrial sectors, while Dominican Republic and Jamaica are moving toward a service dominating economy that is observed in many industrialized countries world wide. How much the trade is important in the economy and what type of commodity is mainly traded determine the rate of urbanization in part. As seen above, in Guyana, sugar cultivation is the main exporting industry and the agricultural trade is almost twice as big as the country's GDP. 6 Given the relative productivity of agricultural and manufacturing sector and the global competitiveness, the cost of migrating from rural agricultural area to the urban industrial area, that is, relatively high rural income for unskilled labor, seems to be too high for families in rural Guyana. The enrollment rate of primary school is maintained at a high level in all of the countries except Haiti, comparing to the average enrollment rates in Sub-Saharan Africa (64%), South Asia (88%) Middle East (89%), or United States (94%). However, the enrollment rate in secondary school is still very low relative to LAC average (65%) or High Income Countries (85%). The enrollment rate in secondary school is exceptionally high in Jamaica, but it does not directly lead to urban migration because the education system is relatively well maintained in rural area. 6 Trade as a percentage of GDP is 198% (World Bank, 2006) 6

In summary, all of the seven countries are at a rather low stage of economic development and their economic profile varies a lot. The composition of agricultural, industrial sector, and export industry have a roll in determining the overall urbanization and the growth of urban population. Previous research investigates the effect of migration on labor force by estimating the probability of employment, which includes a dummy variable indicating whether an individual is a migrant or a nonmigrant (Rives and West, 1992; Spitze, 1984; Lichter, 1980; Sandell 1977). The parameter estimate associated with the migration variable is an estimate of the difference in the probability of employment between migrants and nonmigrants and is assumed to reflect the effect of migration on employment. DaVanzo and Hosek (1981) and Antel (1980) argue, however, that migrants and nonmigrants are self-selected, rather than randomly selected, samples from a population. Therefore, the estimated difference in the probability of employment between migrants and nonmigrants is likely to reflect not only the effect of migration on employment but also differences in the characteristics of migrants and nonmigrants. In order to separate the effects of migration and self-selection bias on the probability of employment, DaVanzo and Hosek (1981) and Antel (1980) argue for including a migration selfselection bias control variable in the model (see Heckman 1979). The parameter estimate associated with the migration self-selection bias control variable measures the effect of having the characteristics of a migrant {nonmigrant), independent of the effect of migrating (not migrating), on the probability of employment. In turn, the parameter estimate associated with migrating (not migrating) in the model is an unbiased measure of the effect of migrating (not migrating) on the probability of employment. The primary reason of this measure of curing self-selection bias is to generate a hypothetical counterpart of a migrant (or nonmigrant). The true effect of migration can be estimated by comparing the current outcomes of migrating to the hypothetically generated outcomes of not-migrating. This type of estimation is needed because the past information of migrants is not often available of the data and the characteristics of migrants before the act of migration cannot be fully controlled. The data used in this paper provides one of the key variable for the past information of migrants, the origin of migrants. The sample cut by origin town allows us to study migrants and nonmigrants who used to live close by before the migrants left the town. This will reduce the geographic heterogeneity of migrants and nonmigrants in the estimation because they are likely to have been under the similar geographical environment. With this information, I simultaneously control the geographic and personal characteristics to reduce the sampleselection bias. I analyze the effect of migration on the education by estimating a logit model of the probability of achieving a certain educational level, controlling the detailed geographical information. Since the education of adults in the household may not be a result of migration but a potential cause of migration, the estimation of education is done only for the appropriate age group. 7 The effect of migration on the income is estimated by a regression of monthly income on personal characteristics. All of these estimations are 7 DaVanzo and Hosek (1981) included the education level to control the characteristic of potential migrants, say, the more people are educated (or skilled), the more likely they will move to urban area. In that case, education is a cause of migration, not the result of migration. To avoid this causality problem, the education estimation in this paper will include the education achievement of young generation of the household, which is less likely be the cause of migration. 7

conducted by origin towns. Definitions of the variables used in this analysis are listed in Table 3. (1) PRIUP (2) SECUP (3) IPCM = = = 0 j 0 j 0 j 1 j 1 j 1 j MIGR MIGR MIGR 2 j 2 j 2 j AGE AGE AGE i 3 j 3 j 3 j NADULT NADULT NADULT 4 j 4 j 4 j NCHILD NCHILD NCHILD 5 j 5 j 5 SEX SEX SEX i ε ω η where, i = j = individual geographical group j The subscript i represents the individual and all the three estimations are conducted separately for group j. Group j represents all individuals who resided in location j in 5 years before the survey is collected. Therefore, every person in group j has the same origin town, but the variable MIGR is equal to 1 for those who migrated within the past 5 years. The dependent variable PRIUP captures the probability of a person to complete primary school or higher educational level. The estimation (1) is done only for individuals with age under 20 to see the effect of migration on the younger generation of the household. SECUP represents completing of secondary or higher education. The second estimation is done only for individuals with age under 20 as well. IPCM is calculated household total income, if available, divided by number of people in the household. If the total income is not available, labor income is used. AGE and SEX are to control individual characteristics. NADULT and NCHILD are included to control household characteristics. Previous studies used marital status in their analyses, but the marital status is excluded in the estimation because of a multicolinearity, that is, a high correlation with number of adults in the household. Table 3 Variable Names and Definitions PRIUP Primary or higher education. Equals 1 if completed. SECUP Secondary or higher education. Equals 1 if completed. IPCM Per capita monthly income MIGR Migration. Equals 1 if migrated to urban area in the past 5 years. NADULT Number of adults in the household NCHILD Number of children in the household 4. Empirical Result Table 4 to Table 10 display the estimation results by country. The regression was run separately by original town. For example in Table 4, each estimation was done by 32 town of origin. I get 32 coefficient estimates associated to MIGRANT, and the column of ESTIMATE represents the average coefficient estimate of those that are statistically significant at. Likewise, STD shows the standard deviation of the coefficient estimates for MIGRANT, only those that are statistically different than zero at the. The third column, " 10% confidence 8

level," shows the percentage of cases, out of 32 cases by origin town, where the coefficient estimates are statistically significant at. Similarly, the fourth column shows the percentage of cases where the estimates are significant at 5% confidence level. Table 4 Estimation Result for Dominican Republic (2002) PRIUP estimate std CONSTANT 0.38 0.03 97.0% 97.0% AGE 0.03 0.00 100.0% 100.0% MIGRANT 0.90 0.05 100.0% 100.0% NADULT 0.08 0.01 100.0% 93.9% NCHILD 0.00 0.00 72.7% 66.7% SEX 0.09 0.01 93.9% 93.9% SECUP estimate std CONSTANT -1.72 0.03 100% 100% AGE 0.01 0.00 97% 97% MIGRANT 0.61 0.03 100% 100% NADULT 0.18 0.00 100% 100% NCHILD -0.21 0.00 100% 100% SEX 0.29 0.01 100% 100% IPCM estimate std CONSTANT 947.05 417.80 61% 55% AGE 225.83 27.65 100% 100% MIGRANT 976.51 369.86 75% 75% NADULT -15.69 79.04 52% 48% NCHILD -31.66 55.26 36% 27% SECUP 2606.68 187.79 94% 94% SEX 946.26 187.73 97% 94% Table 4 shows the estimation result for Dominican Republic. The coefficient estimate for MIGRANT in the first regression of primary education is.89 and it is statistically significant in all towns. The magnitude of the coefficient estimate for MIGRANT dominates coefficient estimates for other variables. The size of effect of other variables is small, but the estimates are statistically meaningful in many cases. The effect of migration also plays a roll in secondary education but the size is a little bit smaller comparing to the primary education. The return for migration is high for the income level in Dominican Republic. Migrant workers earn 976 Dominican Peso which is about US$30 per month on average. The estimate is statistically significant in 75% of towns. Table 5 Estimation Result for Guyana (2002) PRIUP estimate std CONSTANT -1.63 0.16 89% 89% AGE 0.18 0.00 100% 100% MIGRANT 0.74 0.16 90% 90% 9

NADULT -0.53 0.15 63% 63% NCHILD 0.83 0.12 82% 73% SEX 0.03 0.03 82% 73% SECUP estimate std CONSTANT -0.34 0.12 100% 100% AGE 0.00 0.00 82% 82% MIGRANT 0.91 0.11 90% 90% NADULT 0.74 0.10 100% 100% NCHILD -1.28 0.08 100% 100% SEX 0.10 0.02 82% 82% In Guyana, migrant children also have higher probability of finishing primary school. The average coefficient estimate associated to MIGRANT is.74 higher for migrants than nonmigrants. The effect of migration is bigger for the secondary education. The coefficient estimation for MIGRANT from the estimation using SECUP as the dependent variable is.91. This result is significant in 90% of origin towns, that is, 9 towns out of 10 total origin regions. Table 6 Estimation Result for Haiti (2001) PRIUP estimate std CONSTANT -2.26 0.02 100% 100% AGE 0.03 0.00 100% 100% MIGRANT 1.08 0.03 89% 89% NADULT 0.28 0.00 100% 100% NCHILD -0.16 0.00 100% 100% SEX 0.22 0.01 100% 100% SECUP estimate std CONSTANT -2.97 0.03 100% 100% AGE 0.02 0.00 100% 100% MIGRANT 1.36 0.03 100% 100% NADULT 0.29 0.00 100% 100% NCHILD -0.26 0.00 100% 100% SEX -0.26 0.01 100% 100% IPCM estimate std CONSTANT 467.71 3.45 100% 100% AGE 5.52 0.07 90% 90% MIGRANT 668.04 6.01 100% 100% NADULT -7.79 0.57 100% 100% NCHILD -47.82 0.43 100% 100% SECUP 85.39 4.19 100% 100% SEX 10.34 1.61 80% 80% 10

In the first estimation using PRIUP as the dependent variable in Haiti, the average coefficient estimate for MIGRANT is very big, 1.08. This result is statistically significant for 89% of cases, that is, 8 towns out of 9 total origin towns. The average estimate for the coefficient associated to MIGRANT is even higher for the estimation for SECUP. The estimates are statistically significant for all origin towns. In Haiti, the probability of children to finish primary or secondary school is found very high among migrant families rather than non-migrant families. The magnitude of the effect is relatively big comparing to other countries. Members of migrant households earn additional 668 Haitian Gourde, which is about US$15. Table 7 Estimation Result for Honduras (2004) PRIUP estimate std CONSTANT -0.21 0.36 44% 28% AGE 0.06 0.01 100% 100% MIGRANT 0.88 0.23 100% 100% NADULT 1.24 0.42 94% 94% NCHILD 0.03 0.07 33% 17% SEX 0.01 0.16 22% 22% SECUP estimate std CONSTANT -1.89 0.39 100% 89% AGE 0.01 0.01 44% 44% MIGRANT 0.75 0.29 61% 50% NADULT 0.18 0.07 78% 72% NCHILD -0.25 0.06 89% 89% SEX 0.24 0.18 50% 44% IPCM estimate std CONSTANT 959.17 1273.40 47% 35% AGE 46.77 58.95 32% 21% MIGRANT 215.52 1023.19 33% 28% NADULT 3.43 265.03 6% 0% NCHILD -311.86 147.29 83% 72% SECUP 2139.44 648.44 83% 83% SEX 119.83 616.28 11% 6% Table 7 shows the estimation result for Honduras. The estimate of coefficient associated to MIGRANT is.88 for the fist estimation using PRIUP as the dependent variable. The estimates are statistically significant for all 18 origin towns. However, the effect of migration on the probability of completing secondary or higher education is statistically significant in 61% of cases, 11 out of 18 total origin towns with the 10% confidence level. With the, the effect of migration on secondary education completion probability is statistically meaningful only in 9 towns out of 18. The empirical result shows that the effect of migration seems to be vague in the educational achievement in Honduras. 11

The effect of migration also seems vague for per capita income. The additional income for migration is only 215 Honduran Lempira on average that is equivalent to US$10 in 2004. And the result is not statistically significant for most of the cases. Rather, the return for secondary education is very high among other factors. Table 8 Estimation Result for Jamaica (2001) PRIUP estimate std CONSTANT -0.65 0.00 100% 100% AGE 0.14 0.00 100% 100% MIGRANT 0.31 0.00 100% 100% NADULT -0.10 0.00 100% 100% NCHILD 0.04 0.00 100% 100% SEX 0.05 0.00 100% 100% SECUP estimate std CONSTANT -1.06 0.00 100% 100% AGE 0.08 0.00 100% 100% MIGRANT 0.54 0.00 100% 100% NADULT -0.02 0.00 100% 100% NCHILD -0.25 0.00 100% 100% SEX 0.24 0.00 100% 100% As shown in Table 2, Jamaica is one of the urbanized countries and the rate of migration stays very small during the last decade, suggesting that the speed of further urbanization is very low. The estimation result described in Table 8 suggests that they have less incentive to migrate and this is consistent with slow growth of urban population. The average coefficient estimate for MIGRANT is statistically significant for all 14 origin towns, but the magnitude is very small for the both estimation, PRIUP and SECUP. Table 9 Estimation Result for Nicaragua (2001) PRIUP estimate std CONSTANT -0.05 1.16 18% 11% AGE 0.03 0.02 53% 46% MIGRANT 0.47 1.05 51% 48% NADULT 0.00 0.24 17% 12% NCHILD -0.06 0.18 29% 19% SEX 0.05 0.49 5% 3% SECUP estimate std CONSTANT -2.64 1.74 46% 33% AGE 0.01 4.87 27% 20% MIGRANT 1.73 1.10 56% 52% NADULT 0.16 22.41 34% 28% NCHILD 0.04 89.90 48% 39% 12

SEX 0.78 0.65 10% 6% IPCM estimate std CONSTANT 258.52 561.01 71% 66% AGE 7.96 6.81 9% 5% MIGRANT 117.25 219.07 55% 51% NADULT 0.31 97.56 48% 42% NCHILD -107.05 70.58 72% 69% SECUP 354.37 293.27 56% 50% SEX 2.02 181.15 11% 6% Table 9 describes the estimation result for Nicaragua. The survey collects the information of origin towns (where they lived 5 years ago) in very detail, 126 municipals. This allows us to do an empirical study that is close to an ideal case because we can safely assume that there is no geographic heterogeneity in a specific municipal. Therefore, people who lived in a certain municipal 5 years ago are believed to have been under the same geographic environment, and the effect of migration is well captured in the estimation by controlling individual characteristics. The coefficient estimate for MIGRANT in the first regression is.47 on average and they are significant in about half the cases. The reason why we have many insignificant results is partly because the number of observations is too small in many cases. The average coefficient estimate for MIGRANT in the second regression using SECUP is much higher, 1.73, and again, they are statistically significant in half of the cases. The average effect of migration on per capita income is 117 Nicaraguan Cordoba Oro which is about US$5 in 2001. Table 10 Estimation Result for Paraguay (2004) PRIUP estimate std Incidence of significance at Incidence of significance at CONSTANT -0.61 0.02 100% 100% AGE 0.03 0.00 100% 100% MIGRANT 0.66 0.03 100% 100% NADULT 0.18 0.00 100% 100% NCHILD -0.20 0.00 100% 100% SEX 0.00 0.00 82% 71% SECUP estimate std Incidence of significance at Incidence of significance at CONSTANT -1.63 0.02 100% 100% AGE 0.01 0.00 100% 100% MIGRANT 0.60 0.03 100% 100% NADULT 0.23 0.00 100% 100% NCHILD -0.33 0.00 100% 100% SEX 0.02 0.00 81% 81% IPCM estimate std Incidence of significance at Incidence of significance at CONSTANT 601757.77 9719.86 100% 100% AGE 3414.80 202.15 100% 100% 13

MIGRANT 151272.23 7307.55 100% 100% NADULT -31261.96 2160.22 94% 94% NCHILD -66896.78 1786.15 100% 100% SECUP 288113.51 6869.06 100% 100% SEX 10664.26 1197.35 78% 78% As in Table 2, Paraguay is one of the countries where urbanization is at a very high speed. Table 10 shows that migrants have higher probability of completing primary and secondary education. The average estimate for the coefficient associated to MIGRANT is.66 in the first estimation using PRIUP as the dependent variable, and.60 for the second estimation using SECUP as the dependent variable. Notice that we have statistically significant results for all 16 origin towns. The third panel of Table 10 describes that migrants earn 151272 Paraguay Guarani per month in 2004, about US$15. This additional income takes 25% of average income in Paraguay, which suggests migrants to urban area could earn substantial income relative to nonmigrants. 5. Conclusion The empirical results in this paper are summarized in Figure 2. Primary Education is the average estimate for the coefficient associated to the migrant dummy variable in the regression using the completion of primary education as the dependent variable. Secondary Education is the average estimate for the migrants in the regression using the completion of secondary education as the left had side variable. Additional Income is the effect of migration on the per capita income as a share of GDP per capita. Figure 2 Effect of Migration on Education and Income 2.0 1.8 1.6 1.4 1.2 Primary Education 1.0 Secondary Education 0.8 0.6 Additional Income 0.4 0.2 0.0 Dominican Republic Guyana Haiti Honduras Jamaica Nicaragua Paraguay In all seven countries, I found some empirical evidence of higher probability of completing primary education for children of migrant families. The effect is biggest in 14

Haiti and smallest in Jamaica. The result is statistically robust across origin towns of migrants and across countries. For the secondary education, the effect of migration is highest in Nicaragua but not so robust across origin towns. Excluding Nicaragua, the effect is, again, biggest in Haiti and smallest in Jamaica. Therefore, the return to migration in terms of education opportunity for younger generation is arguably highest in Haiti and lowest in Jamaica in the sample countries in this paper. The additional income that migrants earn comparing to nonmigrants is not insignificant. It reaches up to 26% of GDP per capita in Haiti and down to 7% in Nicaragua, and the estimate is fairly robust across origin towns in each country. The magnitude of this income effect as a share of GDP per capita is not completely compatible across countries because the size of national economy varies (Jamaica is seven times richer than Haiti in terns of GDP per capita in Table 2). However, there is no doubt that migrants have a significant amount of income more than nonmigrants even after controlling the geographical and individual characteristics. This paper presents some empirical evidence that migrants achieved their goals to migrate education for their children and higher income for the family. The results in this paper shed light on the assessment of quantity of the education and employment opportunities of migrants. However, it is far from a conclusion that migrants have a better education and better job, which might lead to a higher living standard, because the welfare would rely not only on the quantity but also on the quality of education and employment. Further studies on the quality of jobs and education using micro level data, as in Hannum(1999) and Reardon(2001), are called for a broader understanding of the interaction between family structure, education, labor market and migration. 6. Reference Agesa, Richard U., (2001) "Migration and the Urban to Rural Earnings Difference: A Sample Selection Approach," Economic Development and Cultural Change, Vol 49, No.4, pp.847-865 Antel, J. J. (1980), "Returns to migration: Literature review and critique," Rand Note no. 1480. Cooke, Thomas J., and Adrian J Bailey, (1996) "Family Migration and the Employment of Married Women and Men," Economic Geography, Vol 72 No1, pp. 38-48 DaVanzo, J, and Hosek, J, R, (1981), "Does migration increase wage rates?- An analysis of alternative techniques for measuring wage gains to migration," Rand Note no. 1582. De Jong, Gordon F., (2000) "Expectations, Gender, and Norms in Migration Decision- Making," Population Studies, Vol 54, No.3, pp.307-319 Hannum, E., (1999) "Political Change and the Urban-Rural Gap in Basic Education in China, 1949-1990," Comparative Education Review, Vol. 43, No. 2, pp. 193-211 Lichtet, D. T. (1980), "Household migration and the labor market position of married women," Social Science Research 9, pp.83-97. Mincer, J., (1978) "Family migration decisions," Journal of Political Economy 86, pp.749-73 15

Mohlo, I., (1986) "Theories of migration: A review," Scottish Journal of Political Economy 33, pp.396-419 Reardon, T., (2001) "Rural Nonfarm Employment and Incomes in Latin America: Overview and Policy Implications," World Development, 29 (3) Rives, J. M., and West, J. M. (1992), "Worker relocation costs: The role of wife's labor market behavior," Regional Science Perspectives 223-12. Sandell, S. H., (1977) "Women and the economics of family migration," Review of Economics and Statistics 59, pp. 406-14 Spitze, G, (1984) "The effect of family migration on wives' employment: How long does it last?" Social Science Quarterly 65, pp.21-36. Stark, O., and J. Edward Taylor, (1991) "Migration Incentives, Migration Types: The Role of Relative Deprivation," The Economic Journal, pp.1163-1178 World Bank, (2006) World Development Indicator 16