DOES POST-MIGRATION EDUCATION IMPROVE LABOUR MARKET PERFORMANCE?: Finding from Four Cities in Indonesia i

DOES POST-MIGRATION EDUCATION IMPROVE LABOUR MARKET PERFORMANCE?: Finding from Four Cities in Indonesia i Devanto S. Pratomo Faculty of Economics and Business Brawijaya University Introduction The labour market performance of migrants at destination continues to be a major focus for research on migration. Several studies have indicated that education is one of the crucial aspects for migrants of their lifetime labour market position and their related outcomes at destination (see for examples Betts and Lofstrom, 2000, and Kanas and van Tubergen, 2009). Education would be important not only for international migration but also for rural-urban migrants in the country competing for urban occupations due to the fact that urban areas generally provide more opportunities for the highly educated (Meng, 2001 and Manning and Pratomo, 2013). Most rural-urban migrants moved to the cities for seeking better economic opportunities (work motive) with a substantial pre-migration education from the origin. However, migrants who have received a substantial education prior to migration might also decide to take further education at the time of arrival in order to be easily utilized in their new environment (Kanas and van Tubergen, 2009). It is also possible that migrants who initially aimed at entering the labour market at destination go back to school for further education after a period of employment to advance in their career. On the other hand, migration for education is also common particularly for school-age children moving or family relocating motive (Lucas, 1997). Arendt et al (2012) pointed out that school-age children or younger-arriving migrants are more likely to continue or to complete their education at destination as a natural continuation, taking advantage of better educational environment at destination. 1

Most previous studies on post-migration education focus on the education of immigrants from developing countries to developed countries, given different system and standards of education between developing and developed countries. Using data of immigrants in Netherlands, Kanas and van Tubergen (2009) found that immigrants with post-migration education are paid much higher than migrants without schooling at destination mainly because of the different quality of education between origin and destination. Similar with them, Duleep and Regers (1999) and Arendt et al (2012) showed that immigrants with their home education qualifications found difficulties to be fully accepted in the labour market because of different standard and quality of education between origin and destination. Compared to international migration studies, the role of post-migration education in the case of rural-urban migration particularly in developing countries is a relatively neglected area of research. One possible reason is because of the lack of data for rural-urban migration particularly in the case of developing countries. This study is taking advantage by using a new dataset from Rural-Urban Migration in China and Indonesia (RUMiCI) conducted by the Australian National University focusing specifically on the rural-urban migrants in the four largest recent migrant destination cities in Indonesia including Tangerang, Medan, Samarinda, and Makassar. The other main data set covering migration in Indonesia, the Indonesian Family Life Survey (IFLS), does not have large enough samples for rural-urban migrants, while the Indonesian Population Census does not cover rural-urban migration but migration across districts and provinces. Indonesia, as a large archipelagic country, provides a unique characteristic given the large differences with respect to both quantity and quality of education between urban and rural areas. Compared to rural areas, urban areas in Indonesia are much more supportive at least in terms of facilities and infrastructures of education affecting the difference in outcome of education between urban and rural areas. The differences between urban and rural areas in terms of education will be discussed in more detail in the next section. 2

The first objective of this study is to examine which factors are important in determining the post-migration education among rural-urban migrants in Indonesia. Migrants in the cities arrive with varying levels of education and labour market experiences, depending on their age at the time of migration as well as their household and personal characteristics from the origin. Previous studies showed that the education attainment prior to migration (pre-migration education) is the most important factor in explaining the post-migration education participation. However, the effects of pre-migration education on post-migration education participation tend to be ambiguous. Borjas (1982) and Khan (1997), for example, found that pre-migration education has a negative effect on post-migration education measured by years of schooling at destination, meaning that the higher education attained prior to migration, the smaller probability that migrants will invest more in post-migration education. In other words, pre-migration education acts as a substitute for the benefit of post-migration education (the so-called substitution effect). In contrast, Hansen et al (2001), Cobb-Clark et al (2005), and van Tubergen and de Werfhorst (2007) found that pre-migration education tends to be positively related with the postmigration education investment suggesting a complementary effect between pre- and postmigration education. This effect means that migrants who make large investments of education in the origin tend to continue to make large investments of education at destination. Khan (1997) pointed out that the positive effects of pre-migration education are usually more intense at relatively lower levels of pre-migration education but less intense at higher levels of premigration education. Based on these contradictory previous studies, this study tries to find whether the substitution effect or the complementary effect dominates in the case of ruralurban migrants in Indonesia. The second objective of this study is to investigate whether investing in post-migration education in the cities improves the labour market performances of rural-urban migrants. The labour market performances are measured by the occupational (work) statuses and earnings (wages) at destination. It is predicted that migrants with more schooling after migration (post- 3

migration education) are better off in terms of work statuses and earnings than those who are less or not participating in post-migration education. The section on occupational status modifies a method by Manning and Pratomo (2013) using the multinomial logit model that makes it possible to observe specific effects of the postmigration education on several categories of employment, including formal sector, formalcasual, small-business, and informal sector, going beyond the conventional formal-informal sector ii. Although it is hard to identify which status is the higher or better off one, this study predicts that the formal sector has certainly attracted the better qualified individuals with higher and more stable employment and earnings compared to formal-casual sector and informal sector, while small-business, which is usually included as the informal sector in other studies, generally requires more managerial and specific skills compared to common informal sector (Meng, 2001). Therefore, migrants with more post-migration education are predicted to be more likely employed in the formal sector (or possibly in small-business activities), compared with migrants with less or without post-migration education. Relating to earnings, this study predicts that migrants with more post-migration education will receive higher returns than migrants with less post-migration education, controlling other household and individual characteristics. This section contributes to the methodology by employing the sample selection corrections based on a multinomial logit for a potential selection bias from a non-random sample. This method is used to control for the potential selection bias arising from the correlation between the unobserved factors that affect the choice of occupational statuses (in the multinomial logit model) and the unobserved factors that affect their earnings due to the fact that those individuals who expect higher earnings are more likely to select themselves (self-selection) into the formal sector (as a desirable employment choice), leading to potential sample selection bias in the simple Ordinary Least Square (OLS) estimates. 4

The outline of this paper is as follows. The second section discusses the migration and educational differences among urban and rural areas in Indonesia. In the third section, this paper briefly describes the data and methodology used. This paper then examines the empirical result of the determinants of post-migration education. The discussion then moves to the relationship between post-migration education and the labour market performance measured by occupational statuses and earnings of migrants. The final section provides the conclusions and some implications. Migration and Rural-Urban Educational Differences in Indonesia Rural urban migration has been a feature of the labour market in Indonesia since the late 1960s following the structural transformation process in the economy from the agriculture sector into the modern sectors (Manning, 1987). It is estimated that more than 18 million or close to 15 per cent of total urban population were migrants born in rural areas (Meng and Manning, 2010). The rural-urban migrants are characterized by the domination of young adults seeking for better economic opportunities with a relatively higher level of education than the rural residents who left behind (Speare and Harris, 1986). Many rural residents in Indonesia also moved to the cities to continue their study either because of family relocating or a natural continuation of education. Speare and Harris (1986) showed that many rural residents in Indonesia in 1970s who had completed junior high school moved to the cities due to the fact that the senior high school were not available within commuting distance from their home. Although Indonesia has made improvements in the education sector over the past 30 years, these disparities in terms of the availability of school between urban and rural areas remain marked in the recent times. SMERU (2011) showed that more than 70 per cent of villages (kelurahan) in rural areas in 2008 do not have secondary schools (Table 1). The average distance from the villages in rural areas with no secondary school to the closest school is 8.18 kilometers for junior secondary and 13.16 kilometers for senior 5

secondary. In some remote areas, like in Papua and East Nusa Tenggara, the access is still a problem where students had to walk more than 10 kilometers to go to school. Table 1 Availability and accessibility of schools in urban and rural areas, 2005 and 2008 (%) Villages with no schools (%) Average Distance from Villages with No School 2005 2008 2005 2008 Primary Level Indonesia 10.43 8.14 5.99 5.92 Urban 4.21 3.76 1.09 1.13 Rural 11.76 8.77 6.36 6.12 Junior Secondary Level Indonesia 66.60 58.15 8.97 7.82 Urban 40.76 29.32 1.65 2.13 Rural 72.18 62.05 9.85 8.18 Senior Secondary Level Indonesia 83.78 80.07 14.41 12.51 Urban 54.04 44.04 2.89 3.39 Rural 90.12 84.51 16.04 13.16 Source: Adapted from SMERU (2011). The original source is from Podes, village level data. Although it is hard to provide data comparing the educational quality between urban and rural areas, there is no doubt that the student outcomes and school conditions in rural areas are still lagging behind those in urban areas. This difference stems from the lack of teacher resources and lack of facilities in rural and remote areas affecting the teaching quality and student achievements (Luschei and Zubaidah, 2012). World Bank (2008) and SMERU (2011) showed that the number of primary school teachers was not equally distributed between urban and rural areas. The proportion of schools undersupplied with teacher is much higher in rural and remote areas than in urban areas (Figure 1). The proportion of school undersupplied with teachers is about 37 per cent in rural areas and 66 per cent in remote areas, while the proportion of school undersupplied with teachers in urban areas is 21 per cent. The main reason for this disparity of undersupplied teachers is primarily because of the remoteness of the areas, the lack of 6

adequate housing for teachers, limited transportation to the village, the isolation from family and friends, and limited services and facilities. In contrast, the proportion of schools oversupplied with teachers is 52 per cent in rural areas and 17 per cent in remote areas, compared to 68 per cent of ovesupplied teacher in urban areas (Figure 1). Figure 1 Distribution of Primary School Teachers, Across Urban, Rural, and Remote Areas Source: Adapted from SMERU (2011) and World Bank (2008) Relating to the school facilities, Firman and Tola (2008) reported that schools in rural areas usually lack Information and Communications Technology (ICT) equipment and infrastructure as well as limited staff to manage ICT equipment in schools. In addition, Usman et al (2007) even showed that some rural schools did not have electricity and functioning toilet. There is also evidence that teachers in rural areas tend to have lower education levels than those in urban areas. Based on the survey conducted by World Bank (2008), the proportion of primary school teachers in rural and remote areas holding bachelor degree (and more) is 17.4 per cent and 5.2 per cent respectively, compared to 27 per cent of teachers in urban areas. Luschei and Zubaidah (2012) also found that teachers in rural Indonesia have little access to 7

training opportunities and to assess their professional needs compared to teachers in urban areas because of large distances between their school location and teacher training institution. Given the differences in both quantity and quality aspects of education between urban and rural areas in Indonesia, the role of post-migration education might become important for migrants for surviving in urban jobs. Although many migrants arrived in the cities with relatively high levels of education, many of them remain actively participating in school in order to compete for more stable employment sectors at destination. According to Hill and Thee (2012), the elite or good reputation high schools and universities in Indonesia are mainly also located in urban areas. However, whether migrants participate in post-migration education strongly depends on many factors including the individual and household characteristics. The costs of education investment are also generally higher in urban areas than in rural areas. Data and Methods The main source of the data used in this study is the Rural-Urban Migration in China and Indonesia (RUMiCI) survey conducted by the Australian National University. RUMiCI is a longitudinal household level survey conducted in China and Indonesia from 2008 to 2011 to investigate the labour market activities and welfare of individuals who have migrated from rural to urban areas. Using the RUMiCI data set for Indonesia, this study employs a sample that contains rural-urban migrants who were living in the four largest recent migrant destination cities in Indonesia including Tangerang (sub-urban part of Jakarta), Medan, Samarinda, and Makassar, representing the four broad geographic regions in Indonesia: Java, Sumatera, Kalimantan, and Sulawesi. To distinguish urban and rural areas, RUMiCI follows the Indonesian statistical office (BPS) classification based on the population density, number of agricultural households and presence of some typical urban infrastructures. This study firstly examines the determinants of post-migration education among rural-urban migrants in Indonesia. The definition of rural-urban migrants used in this study is slightly 8

different from that of other studies of the RUMiCI data set for Indonesia (see some studies using Indonesian RUMiCI data set in Meng and Manning, 2010). To meet the definition of ruralurban migrant, this study examines migrants who have lived for at least five years continuously in a rural area and who arrived in the cities when they were older than 15 (the age after graduating from junior secondary school). These conditions ensure that most of the migrants finished primary and junior secondary education before migration. In Indonesia, primary and junior secondary educations are compulsory by regulation. Therefore, migrants that participate in post-migration education in this sample are more likely to do so optionally, rather than being compelled to do so. The other reason is because individuals who were living less than five years in a rural area or who were migrating at the childhood ages (age of primary and junior secondary schools) tend to easily adapt to the environment in urban areas and will be indistinguishable in skills and experience with non-migrants (Resosudarmo et al, 2010). Following Khan (1997) and Cobb-Clark et al (2005), this study uses two measures of postmigration education as dependent variables, i.e.: (1) a dummy variable measuring whether migrants enroll in school after migration (estimated using a binary probit), and (2) years of schooling obtained after migration (estimated using OLS). In the RUMiCI 2011, migrants were asked detailed questions on their educational attainment before migration to the cities (premigration education), while in RUMiCI 2009-11, during the survey conducted in the cities, migrants were asked about the highest education completed. If the highest education completed of migrants during the survey conducted is higher than their pre-migration education, we can say that migrants attended school after their arrival in the cities (participating in post-migration education) iii. In the second measure, the years of schooling obtained after migration is calculated by differentiating the total years of schooling obtained during the survey conducted and the years of schooling before migration. The explanatory variables include several individual and household characteristics as follows: pre-migration education (measured by years of schooling before migration), age at migration, 9

years since migration (duration of migration), dummy of gender (males and females), dummy of marital status before migration (married and unmarried/singles), dummies of motives of migration (work, school and other motives), dummies of location of residence/cities (Medan, Samarinda, Makassar, and Tangerang). The variables of pre-migration education, age at migration, and years since migration include the quadratic terms allowing potential non-linear relationship with post-migration education, as suggested by previous studies (see for examples Khan, 1997 and van Tubergen and de Werfhorst, 2007). In addition, financial resources of migrants prior to migration potentially also work as an important contributor for the post-migration education decision. Unfortunately, the household s financial information, such as earnings, expenditure, and assets owned by the household prior to migration is not available from the survey. The survey provides only information on financial resources of households during the survey conducted (after migration) but it will give a complicated causality effects if it used a proxy. This study therefore uses the estimated earnings in rural areas as an additional explanatory variable as a proxy for household s financial resources prior to migration iv. Although it is not a perfect measure, this variable at least provides the potential income if they are living in rural areas. Table 2 presents descriptive statistics for the main variables of post-migration education determinants across rural-urban migrants. As presented in table 2, 21.9 per cent of migrants in the sample were enrolled in school after migration, while most of them attended no school after migration. It means that not every migrant enrolls in education at destination, but most of them were educated at the origin. The average years of schooling of migrants after migration are 0.84 years. The number is low because most of the sample attended no school after migration. The average years of schooling before migration (pre-migration education) is 9.6 years indicating that migrants on average move to the cities with at least junior secondary 10

education. Most of the migrants are young when they move to the cities at around 22 years old and only 37.7 per cent of migrants are married before migration. Comparing gender, the proportion of male migrants is relatively similar to female migrants. More than 50 per cent of migrants move to the cities for work, while the proportion of migrants who migrate for education (school motive) is only 9 per cent. Table 2 Descriptive Statistics for Determinants of Post-Migration Education Mean S.D. Dependent Variables Enrolled in School 0.219 0.414 Years of Schooling 0.840 1.889 Independent Variables Pre-Migration Education (years) 9.601 3.241 Pre-Migration Education Sq. 102.668 56.310 Age at Migration 21.937 6.844 Age at Migration Sq. 528.019 441.182 Years Since Migration 19.106 11.644 Years Since Migration Sq. 500.463 559.331 Males 0.553 0.497 Married Before Migration 0.377 0.485 Work Motive 0.565 0.496 School Motive 0.091 0.287 Medan 0.214 0.411 Samarinda 0.194 0.395 Makassar 0.190 0.392 Log of Earnings in Rural Areas 13.217 0.823 Source: RUMiCI Survey Secondly, this study examines the effects of post-migration education on the labour market performances of migrants measured by occupational status using a multinomial logit model. The analysis is restricted to migrants who have lived for at least five years continuously in a rural area, who arrived in the cities when they were older than 15, and were no older than 64 during the survey, based on the presumption that migrants older than 64 have left the labour 11

market. This study follows Manning and Pratomo (2013) approach, expanding the conventional definitions of the formal and informal sectors by introducing formal-casual and small-business as different employment positions in the urban labour market. Formal-casual, in practice, is a category at the same environment with the formal sector, but usually does not benefit from permanent job security, health insurance, and old-age pensions in the formal sector, while small-business is a category quite close to the informal sector but requires more managerial and other specific skills (Meng, 2001). As a result, four choices of occupational status used as dependent variables for the multinomial logit model including: (1) formal sector, (2) formal-casual, (3) small-business, and (4) informal sector. The unemployed are excluded from the analysis due to the relatively limited sample of unemployed. The formal sector employment is the highest quality of employment and has certainly attracted the best qualified individuals with higher and more stable employment and earnings. In this study, formal sector employment is defined specifically as regular employees in the firms with five workers or more; wage workers in the public sector; or self-employed with relatively big assets (more than Rp.100 million or approximately $10,000 at 2013 exchange rates) or employing more than 20 workers. In contrast, the formal-casual sector is composed of casual or contract workers (not permanent workers) in establishments with more than five workers and all regular employees in small establishments with less than five workers. Manning and Pratomo (2013) showed that formal-casual usually also acts as a temporary employment opportunity before migrants move to the formal sector in the long-term. Furthermore, the small business sector is composed of employers or self-employed with productive assets between Rp.5 million and Rp.100 million, but employing no more than 20 workers. Finally, the informal sector covers self-employed with assets valued at less than Rp.5 12

million and employing no more than 20 workers. In addition, unpaid family workers are also included in the informal sector. The work status classification used in this study is presented in table 3 combining type of employment, number of workers, and total productive assets. Table 3 Work Status Definition Number of Workers Assets (<5) (5-19) (>20) (Rp million) Private Sector Formal Casual Formal Formal Any Casual Workers Informal Formal-Casual Formal-Casual Any Employer/ Informal Small Business Formal <5 Self-Employed Small Business Small Business Formal 5-100 Formal Formal Formal >100 Public Sector Formal Formal Formal Any Unpaid Family Workers Informal Informal Informal - The main explanatory variable is the post-migration education, measured by the number of years migrants have followed education after migration. The other explanatory variables used in the multinomial logit model include: Pre-migration education (measured by years of schooling before migration), years since migration (duration of migration), age, dummy of gender (males and females), dummy of marital status (married and not-married/singles), dummies of household status (household head, spouse, and others), log of financial assets owned by the household, dummy of experience in non-agricultural sector before migration (non-agricultural experience and agricultural experience), dummies of type of jobs (professional, middle-class workers, and unskilled workers) dummies of location of residence/cities (Medan, Samarinda, Makassar, and Tangerang). 13

Similar to above, the quadratic terms of some explanatory variables, including pre- and post- migration education, year since migration and age are also added, allowing potential nonlinearity with the occupational statuses. Table 4 Descriptive Statistics for Occupational Statuses Formal Sector Formal-Casual Small Business Informal Sector Mean S.D. Mean S.D. Mean S.D. Mean S.D. Post-Migration Educ. 1.078 1.928 0.655 1.821 0.669 1.723 0.885 2.027 Post-Migration Educ. Sq. 4.869 11.541 3.729 18.438 3.394 12.657 4.877 16.391 Pre-Migration Educ. 11.124 2.525 9.616 2.899 9.142 3.036 8.235 3.412 Pre-Migration Educ. Sq. 132.334 53.704 100.825 50.067 92.717 49.540 79.404 53.457 Years Since Migration 20.289 9.580 17.181 9.916 22.244 9.852 18.896 10.304 Years Since Migr. Sq. 503.173 408.551 392.955 404.028 591.095 467.740 462.819 465.144 Age 40.851 9.949 38.395 10.159 45.259 9.582 42.073 10.307 Age Squared 1767.478 819.505 1576.836 825.560 2139.543 855.857 1875.973 885.895 Males 0.734 0.442 0.689 0.464 0.677 0.469 0.504 0.501 Married 0.859 0.348 0.842 0.366 0.906 0.294 0.892 0.311 Household Head 0.779 0.415 0.706 0.457 0.717 0.452 0.546 0.499 Spouse 0.191 0.394 0.220 0.416 0.276 0.449 0.427 0.496 Log of Assets 17.659 1.909 16.774 2.089 18.324 1.581 16.879 2.164 Non Agricultural Experi. 0.400 0.491 0.367 0.483 0.409 0.494 0.465 0.500 Professional 0.328 0.470 0.136 0.343 0.291 0.456 0.142 0.350 Intermediate Workers 0.445 0.497 0.362 0.482 0.299 0.460 0.231 0.422 Medan 0.245 0.431 0.181 0.386 0.331 0.472 0.285 0.452 Samarinda 0.182 0.386 0.249 0.433 0.307 0.463 0.162 0.369 Makassar 0.860 0.391 0.153 0.361 0.157 0.366 0.169 0.376 Ln Wage* 14.391 0.627 13.975 0.625 14.484 0.838 13.775 0.986 N 335 177 127 260 Source: RUMiCI Survey Note: * used in Wage Equation Table 4 presents the descriptive statistics of the variables used in the multinomial logit model. The average years of schooling of migrants before and after migration (pre- and post-migration education) for formal sector employment are higher than migrants in the other sectors (at 11.1 and 1.1 years respectively), suggesting that the formal sector attracts better educated workers than the other sectors. Interestingly, the post-migration education for informal sector 14

employment is a bit higher than those in the small business and formal-casual sectors, although these two sectors have lowest pre-migration years of schooling. As indicated by years since migration in table 4, an individual with longer experience living in the cities tends to work in small business and the formal sector (the averages are 22 and 20 years, respectively), while individuals working in formal-casual and informal sectors have been living in the cities around 17-18 years. The proportions of males are higher in all occupational statuses, although the proportion of females is relatively high in the informal sector which includes the unpaid family workers category. Most individuals are married and the proportion is relatively similar for all categories. The household head predominantly works in the formal, formal-casual, and small business sectors, while his/her spouse tends to work in the informal sector. Household assets is an important contributor for small business category, but less likely to be important for formal-casual and informal sectors. The labour market performance is then measured by wages received related to the occupation status across different post-migration education attainment categories. There is a potential problem of selection bias relating to the appropriate specification in this earnings estimation as individuals in the sample might select themselves (self-selection) into an employment sector (or category) where they have a preference depending on the level of wage offers. Therefore, they are likely to be non-random samples within the population. This implies that the unobserved factors which affect the choice of employment sector are also likely to be correlated with the unobserved factors in the wage equation, suggesting a potential sample selection bias using Ordinary Least Square (OLS) estimator. To control for a potential sample selection bias, a two-step procedure of Lee s selection biased corrections based on the multinomial logit will be employed when estimating the wage equation (Lee, 1983). 15

Specifically, the two-step procedure of selection biased corrections used in this study is as follow: * (1) ys z s s ; s = 1, 2, 3, 4 (2) ws xs ms s ; The first-stage of estimation is actually the multinomial logit model used for the occupational status equation above where the dependent variable (y s ) is a categorical variable representing four different employment categories, including (1) formal sector, (2) formal-casual, (3) smallbusiness, and (4) informal sector. The multinomial logit model of these four categories of employment is estimated in order to obtain the predicted values used to generate the selectivity term associated with occupational status category. The second-stage of estimation is the wage equation including the selectivity term result (m s ) from the first-stage estimation. The log of financial assets owned by the household is used as an identifying variable, so it is excluded from the second-stage of estimation. Similar to the above, the main explanatory variable is the post-migration education as measured by post-migration years of schooling, while the other explanatory variables are broadly the same as the variables used in the multinomial logit estimate. Empirical Results Table 5 presents the findings of post-migration education determinants among rural-urban migrants in four Indonesian cities. In the first column, a binary probit model of whether the migrants have enrolled in school after their arrival in the cities is estimated. Pre-migration education measured by years of schooling prior to migration is negatively related to the enrollment, suggesting that migrants with higher level of education before migration are less likely participating in school after migration. Therefore, we can conclude that the substitution effect dominates the complementary effect. This finding is similar to Borjas (1982) and Khan (1997) indicating that pre-migration education substituted for the benefit of post-migration education. 16

Although the pre-migration education effect is negative, the pre-migration education squared variable shows positive and significant effect suggesting that overall pre-migration education has a non-linear relationship. In this case, the positive effect is only found among migrants with an advanced education prior to migration with the turning point at around 12 years of schooling (senior secondary education level) v. This is consistent with Hill and Thee (2012) discussion that many migrants with senior secondary education continue their post-secondary education in the cities as the universities and colleges are concentrated in urban areas. However, the positive effect of pre-migration education squared is relatively small compared to the negative effect of pre-migration education, suggesting that the substitution effect remains dominant compared with the complementary effect. Table 5 Determinants of Post-Migration Education Investment Probit Analysis: OLS Analysis Dependent Var: Enrolled in School Dependent Var: Post-Migration Years of Schooling (1) (2) Coef. P value Coef. P value Pre-Migration Education -0.392 0.000-0.695 0.000 Pre-Migration Education Sq. 0.016 0.000 0.029 0.000 Age at Migration -0.087 0.004-0.126 0.000 Age at Migration Sq. 0.001 0.006 0.002 0.000 Years Since Migration 0.041 0.007 0.019 0.192 Years Since Migration Sq. -0.000 0.073-0.000 0.603 Males 0.193 0.086 0.256 0.020 Married Before Migration -0.309 0.007-0.466 0.000 Work Motive -0.206 0.088-0.148 0.213 School Motive 1.164 0.000 1.902 0.000 Medan 0.043 0.772-0.101 0.497 Samarinda 0.339 0.017 0.316 0.026 Makassar 0.607 0.000 0.632 0.000 Earnings in Rural Areas 0.203 0.003 0.288 0.000 Constant -0.892 0.360 1.977 0.049 N 1027 1027 R-squared 0.234 0.310 17

Consistent with the binary probit findings, in the second column of table 5, using the postmigration years of schooling as a dependent variable, the pre-migration education also shows a non-linear relationship. Similarly, the negative effect (substitution effect) is found particularly at the lower level of education, while at advanced levels, the complementary effect dominates the substitution effect with the turning point also found at around 12 years of schooling. Using immigrants in the US data, Khan (1997) also showed that pre-migration schooling up to the secondary level (12 year) is found to have a substitution effect with US schooling, while postsecondary schooling complements for migrants the secondary level education at origin. Another important factor influencing the post-migration education investment is the age at migration. Previous studies showed that the age at migration is more likely to be negatively related to the post-migration education as younger-arriving migrants have lower opportunity costs and a longer remaining working life at destination (Khan, 1997 and Betts and Lofstrom, 2000). In other words, migrating at older ages leads to less post-migration education. Similar to pre-migration education variable, the non-linear relationship between age at migration and post-migration education is also significant in this study. For migrants who arrive at relatively younger ages, this study supports the previous studies that the age at migration has a negative effect on post-migration education. This effect is interestingly different among the older-arriving groups of migrants, as indicated by the positive effect of the pre-migration education squared variable. Although the marginal effect is significant, it is relatively small, suggesting that this positive effect is limited. The turning point is found at 34-35 years old for both specifications. This specific age potentially indicates a positive effect driven by migrants migrating at older ages who are going back to school for further education after a period of employment to advance in their career. However, it should be further explored. 18

The year since migration variable measuring duration of stay in the cities is positive and significant in the enrollment specification, but it is not significant on post-migration years of schooling. It means that duration of stay in the cities only affects the enrollment or the participation in school in the cities but it does not significantly increase the years of schooling. Comparing gender, male migrants participate in around 2 years more of post-migration schooling than female migrants, but the difference according to whether they enrolled in school after their arrival is not significant at the 5 per cent level. Whether married before migration, measuring the role of partner, has negatively significant effects on post-migration education for both enrollment and years of schooling specifications. In other words, if migrants married prior to migration, they are less likely to participate in education upon arrival in the cities or have lower post-migration years of schooling. Alternatively, we can also say that the post-migration education is dominated by migrants arriving as singles. Similar to van Tubergen and de Werfhorst (2007), migrants who migrate for education are those most likely to participate in post-migration education. The finding also shows that they tend to have longer post-migration years of schooling than migrants with other migration motives. In contrast, migrants with work motive are not significant on post-migration education. As mentioned by Tubergen and de Werfhorst (2007), labour migrants are less likely to participate in post-migration education as they move mainly for economic reasons, but in the case of rural-urban migrants in Indonesia it is also common for migrants to go back to school for further education after a period of employment. The city of residence also affects migrant decision to participate in education and the postmigration years of schooling. Migrants in Samarinda and Makassar are more likely to participate in education and participate in more years of schooling after migration than migrants in Tangerang (the reference). Tangerang is well-known as an industrial city and is dominated by labour migrants. 19

Finally, the estimated earnings in rural areas as a proxy for financial resources prior to migration are positively related with the probability of being enrolled in school after migration and the duration of post-migration education. Although it is not a perfect proxy for financial resources prior to migration, we can generally say that financial resources at the origin work as an important factor to an individual s ability to participate in education. In other words, migrants with less financial resources at the origin are less likely to invest in post-migration education. Table 6 presents the findings of the multinomial logit regression across the occupational statuses measuring the labour market performance of migrants. The finding significantly shows that an increase in post-migration years of schooling increases the probability of migrants being employed in the formal sector, the more stable occupational status (marginal effect is 0.084). However, using post-migration education squared, the marginal effect is not significant at the 5 per cent level. On the other hand, an increase in post-migration years of schooling decreases the probability of being employed in formal-casual (marginal effect is -0.039) and small business activities (marginal effect is -0.029), while the effect of post-migration education on the probability of being employed in the informal sector is not significant. In other words, this study suggests a potential movement for migrants with post-migration education from formal-casual or small business activities into the more desirable sector, the formal sector. Compared to post-migration education, pre-migration education is not significant across occupational statuses suggesting that pre-migration education does not seem to be an advantage to be employed particularly in the formal sector. However, using quadratic term, the coefficient of pre-migration education squared is positive and significant in the formal sector, suggesting that only for migrants with an advanced degree (higher average years of schooling) from home is there a significant and positive relationship to the probability of being employed in the formal sector. Stated differently, we can say that it is difficult for migrants with a lower 20

education level from home and without post-migration education to be employed in the formal sector. Although this study found that migrants with advanced (higher) labour market quality in both pre- and post-migration are more likely to be employed in the formal sector, comparing the marginal effects of pre- and post-migration education this study suggests that post-migration education has a larger effect on the probability of being employed in the formal sector. This study then concludes that migrants with more post-migration education are significantly better off than those with more pre-migration education. Supporting this finding, using data from immigrants in Netherlands, Kanas and van Tubergen (2009) also found that the returns to origin-country education are lower than to host-country education. Table 6 Multinomial Logit for Work Status Formal Sector Formal-Casual Small Business Informal Sector Coef. P value Coef. P value Coef. P value Coef. P value Post-Migration Educ 0.084 0.001-0.039 0.037-0.029 0.029-0.015 0.445 Post-Migration Educ. Sq. -0.007 0.071 0.005 0.052 0.002 0.333 0.001 0.809 Pre-Migration Educ. 0.010 0.767 0.021 0.434 0.007 0.713-0.037 0.156 Pre-Migration Educ Sq. 0.003 0.066-0.002 0.283-0.001 0.217 0.000 0.804 Years Since Migration 0.032 0.012-0.010 0.227-0.004 0.489-0.017 0.055 Years Since Migration Sq. -0.000 0.067 0.000 0.150 0.000 0.907 0.000 0.326 Age -0.032 0.190 0.013 0.460 0.007 0.570 0.012 0.532 Age Squared 0.000 0.331 0.000 0.310 0.000 0.777 0.000 0.925 Males 0.064 0.417-0.037 0.571 0.056 0.284-0.083 0.298 Married -0.051 0.497 0.051 0.304-0.029 0.610 0.028 0.668 Household Head 0.108 0.300-0.183 0.034 0.135 0.091-0.060 0.574 Spouse -0.089 0.549-0.229 0.000 0.273 0.208 0.045 0.760 Log of Assets 0.005 0.671-0.022 0.010 0.043 0.000-0.026 0.009 Non Agriculture Exp. -0.057 0.137-0.048 0.119 0.001 0.957 0.104 0.005 Medan -0.115 0.014-0.058 0.137 0.103 0.014 0.071 0.149 Samarinda -0.081 0.096 0.054 0.222 0.110 0.014-0.084 0.066 Makasar -0.154 0.002-0.016 0.730 0.059 0.214 0.112 0.061 N 899 Pseudo R-sq 0.149 21

Duration of stay in the cities (years since migration) is important to increase the probability of being employed in the formal sector (the marginal effect is 0.032). It means that the longer the experience of living in the cities, the more likely if is that they will be employed in the formal sector. The years since migration squared is not significant. On the other hand, the number of years since migration has a negative impact on the probability of being employed in the informal sector (marginal effect is -0.017), suggesting a potential movement from informal into formal sectors for migrants with longer duration of stay in the cities. The amount of financial assets owned by the household head is likely to affect migrants to open a small business in the cities (the marginal effect is 0.043), but less likely important for formalcasual and informal sectors. Interestingly, non-agricultural experience from the pre-migration experience significantly affects the probability of being employed in the informal sector upon arrival, but there is no significant impact on the probability of being employed in the formal sector. It seems that the most common type of occupation of migrants working prior to migration is also in the informal sector, due to the limited professional jobs in the formal sector in rural areas. Migrants in Medan, Samarinda, and Makassar are less likely to be employed in the formal sector compared to Tangerang. As mentioned before, Tangerang is a popular industrial and manufacturing city, an integral part of the mega-urban region of Jakarta, with large opportunities to work in the formal sector. In contrast, small business seems to grow more in Medan and Samarinda. In the final section, wage determination related to the occupational status measuring another labour market performance is estimated using Lee s selection biased corrections. As presented in Table 7, the selectivity term is significant, suggesting a possible biased result using OLS estimator vi. This analysis focuses only on wages in formal sector employment because the preand post-migration education variables are not significant in all other sectors and wages tend to be more variable in other sectors (the results are available upon request). 22

Consistent with the multinomial logit finding for the occupational status, the post-migration education is positively related to the wage received by migrants in the formal sector. Based on the result, a one year increase in post-migration education increases wages of workers in the formal sector by 3.7 per cent. This result suggests that migrants who participate in postmigration education not only gained from the higher possibility to be employed in the formal sector but also received greater returns to post-migration education. Table 7 Wage Equation for Work Status in the Formal Sector Lee's Selection (1) OLS (2) Coef. P value Coef. P value Post-Migration Educ 0.376 0.000 0.074 0.084 Post-Migration Educ. Sq. -0.027 0.001-0.004 0.550 Pre-Migration Educ. 0.122 0.082-0.062 0.291 Pre-Migration Educ Sq. 0.011 0.000 0.006 0.022 Years Since Migration 0.116 0.001 0.005 0.853 Years Since Migration Sq. -0.001 0.046 0.000 0.525 Age -0.065 0.218 0.035 0.477 Age Squared 0.000 0.538 0.000 0.423 Males 0.410 0.001 0.253 0.044 Household Head 0.477 0.019 0.092 0.629 Spouse 0.109 0.636 0.232 0.324 Non Agriculture Exp. -0.065 0.380 0.093 0.166 Medan -0.308 0.018 0.133 0.130 Samarinda -0.330 0.007 0.038 0.678 Makasar -0.536 0.001 0.023 0.816 Selectivity term -2.212 0.000 Constant 9.333 0.000 12.904 0.000 N 334 334 Pseudo R-sq 0.299 0.253 However, interestingly, there is also a tendency for non-linear relationship suggesting that the wages decrease among migrants with advanced post-migration education. A possible explanation is that a higher post-migration education does not always bring higher income for 23

migrants with advanced level of education. The turning point is at 5.7 years, so it is around the levels of post-graduate if migrants for example begin their post-migration education at undergraduate level. It is possible that migrants with undergraduate degree but more experience are paid more than those with a recent post-graduate degree but less experience. However, this should be further explored. In contrast, the effect of pre-migration education is positive. The pre-migration education squared also shows a positive effect, indicating no evidence of non-linearity in the case of premigration education. In other words, we can say that an increase in pre-migration years of schooling will absolutely increase their earnings in the cities. Comparing the coefficients of preand post-migration education, the effect of pre- migration education is much smaller, suggesting that post-migration education has a stronger effect on the returns to education than pre-migration education. This finding is similar to Betts and Lofstrom (2000) finding that postmigration education is found to have larger effect on wages than if education is obtained before migration. Looking at the other explanatory variables, years since migration shows a positive relationship, suggesting that the longer the migrants have moved to the cities, the more likely they will be paid more in the formal sector. However, the years since migration squared is negative, suggesting that long-term migrants tend to be paid lower than recent migrants. This is possible due to migrants with longer duration of stay usually being outside their productive ages during the survey conducted. Comparing gender, differences between males and females remained considerable. Male migrants are paid more than female migrants in the formal sector. Non-agricultural experience does not seem to affect the wage in the formal sector as most of their experiences are in the informal sector (see above). Finally, migrants in Medan, Samarinda, and Makassar, cities outside Java, are generally paid less than migrants in Tangerang due to the fact that the 24

minimum wage policies covering the formal sector are also higher in Tangerang than the other cities. Conclusions This study examines the effects of post-migration education on the labour market performances among rural-urban migrants in four Indonesian cities using RUMiCI data set for Indonesia. The main finding of this study is that post-migration education contributes significantly to the labour market performance in terms of work status and wages, compared to pre-migration education. In terms of work status, migrants with more post-migration education are more likely to be employed in the formal sector compared to migrants with less or no postmigration education. In contrast, education seems to be less important for migrants in the informal sector. Relating to earnings, migrants with more post-migration education also tend to be paid more than those migrants with less or no post-migration education. In addition, premigration education tends to be negatively related to post-migration education suggesting a dominance of substitution effect. One implication of the study is that education in both pre- and post- migration stages is an important factor for rural-urban migrants to survive in desirable urban environment. Urban areas generally provide more opportunities for the highly educated. For migrants with less education in both pre- and post-migration, they are less likely to be employed in the formal sector with relatively higher earnings and more likely to be employed in formal-casual or informal sector activities. What is not clear from this analysis is whether the importance of post-migration education is influenced by the differences in quality between urban and rural areas or simply because of limited infrastructures of advanced level of education in rural areas. This issue is beyond the scope of this analysis. Although this study found that post-migration education is important for migrants in terms of their labour market performances in the cities, this study does not examine the importance of sheepskin effects, the effect of holding a specific education level 25