DISCUSSION PAPER SERIES. Schooling Forsaken: Education and Migration. IZA DP No Ilhom Abdulloev Gil S. Epstein Ira N. Gang

Transcription

1 DISCUSSION PAPER SERIES IZA DP No Schooling Forsaken: Education and Migration Ilhom Abdulloev Gil S. Estein Ira N. Gang JANUARY 2019

2 DISCUSSION PAPER SERIES IZA DP No Schooling Forsaken: Education and Migration Ilhom Abdulloev Oen Society Institute Assistance Foundation in Tajikistan and IZA Gil S. Estein Bar-Ilan University, CReAM and IZA Ira N. Gang Rutgers University, CReAM, IOS Regensburg, IZA and RWI JANUARY 2019 Any oinions exressed in this aer are those of the author(s) and not those of IZA. Research ublished in this series may include vies on olicy, but IZA takes no institutional olicy ositions. The IZA research netork is committed to the IZA Guiding Princiles of Research Integrity. The IZA Institute of Labor Economics is an indeendent economic research institute that conducts research in labor economics and offers evidence-based olicy advice on labor market issues. Suorted by the Deutsche Post Foundation, IZA runs the orld s largest netork of economists, hose research aims to rovide ansers to the global labor market challenges of our time. Our key objective is to build bridges beteen academic research, olicymakers and society. IZA Discussion Paers often reresent reliminary ork and are circulated to encourage discussion. Citation of such a aer should account for its rovisional character. A revised version may be available directly from the author. Schaumburg-Lie-Straße Bonn, Germany IZA Institute of Labor Economics Phone: ublications@iza.org.iza.org

3 IZA DP No JANUARY 2019 ABSTRACT Schooling Forsaken: Education and Migration We examine the henomenon of forsaken schooling resulting from oortunities abroad. The brain-drain/gain literature takes as its starting oint the migration of educated/ rofessional labor from oor origin countries to richer host countries. While high-skilled migration is orrisome, many international migrants accet lo-skilled ositions in host countries. Their illingness to do so arises from very large host-home earnings differentials. At home this can lead to reduced educational investment as eole forgo schooling because of oortunities to migrate to high aying lo-skilled jobs. This suggests ossible time-inconsistencies beteen short-run economic gains from migration and negative longterm effects from missing human-caital investment. We analyze data from Tajikistan, here aroximately one-third of the labor force orks outside of the country. We offer an exlanation of our emirical results ith a theoretical model, alloing us to establish the circumstances under hich this tye of forsaken schooling can occur and the trade-offs that olicymakers need consider. JEL Classification: Keyords: O15, P46, F22, I24 migration, tras, overty, inequality, education, skill Corresonding author: Ilhom Abdulloev Oen Society Institute Assistance Foundation (OSIAF) 37/1 Bokhtar St. Dushanbe, Tajikistan ailhom@hotmail.com We thank Barbara Dietz, Ksenia Gatskova and Melanie Khamis for long conversations on this material, and IOS-Regensburg for hel ith the data. We also thank seminar and conference articiants for their comments at the German Develoment Economics Conference, Heidelberg June 2016, ACES 2016, IOS-Regensburg 2nd Annual Conference in July 2015, RWI-Essen in June 2015, and IZA s Annual Migration Meeting in May 2013 (a much earlier incarnation of the ideas in this aer). Part of this ork as comleted hile Ira Gang as visiting the research institute India Develoment Foundation, Gurgaon, India. He thanks them for their courtesy and discussions. This aer does not necessary reflect IDF s or OSIAF s vies and olicies.

4 1. Introduction The ersistence of large age differentials among countries continues to have dramatic influences on eole s lives. In examining international migration, the literature offers to oosing vies of its effects on the sending economy: the brain-drain and the brain-gain. 1 The brain-drain highlights the negative direct imacts of skilled/educated emigration on those left behind; not necessarily the immediate family for hom there is an immediate and obvious trade-off, but the hole economy (Bhagati and Hamada (1974); Grubel and Scott (1966)). The brain-gain emhasizes the second-round ositive imacts on the source country of international emigration through remittances, enhanced returned migrants skills (Co, Gang and Yun, 2000; Dai, Liu, and Xie, 2015) and skill acquisition by those intending to migrate (Beine, Docquier and Raoort, 2008). Remittances alleviate financial constraints faced by migrants' families in the home country and may be used to enhance educational attainment by migrants' children or others in the household. With increased income, migrant families can afford to ay school fees, transortation and school essentials. In addition, families can hire labor to ork in households, family oned businesses and farms, thereby freeing children from doing such ork and alloing them to send more time on education (Dimova, Estein and Gang, 2015). Duryea, Cox and Ureta (2003) find remittances have a significantly larger imact than other income on school retention. Their results are consistent ith subsequent findings by others (Acosta, 2006; Calero, Bedi, & Sarro, 2009). Amuedo-Dorantes and Pozo (2010) find a ositive effect of remittances on children's education in the Dominican Reublic. Yang (2005) finds that exogenous shocks to foreign exchange rates, increasing the value of remittances received by migrant families, had favorable effects on the educational attainment of children. Another grou of brain-gain studies argue that the groth of a migration culture might induce more human caital formation in migrants' home country (Mountford, 1997; Stark, Helmenstein, and Prskaetz, 1997, 1998; Stark and Wang, 2002; Edards and Ureta, 2003; Piracha, Randazzo and Vadean, 2013). Beine, Docquier and Raoort (2001, 2008) suggest that migration might lead to a "beneficial brain-drain" through a demonstration effect if the benefits of increased investments in education exceeds the costs of skilled labor emigration. On the fli side, as often migration results in immigrants orking at lesser skilled jobs then 1 See Dimova (2017) for an excellent ersective on this literature. Page 2 of 34

5 their home country training ould suggest, there is also discussion of a brain-aste henomenon among immigrants (Weiss, Sauer and Gotlibovski, 2003; Özden, 2006). We argue that given existing international age differentials and information on the earnings of lo skilled labor in higher income host countries, individuals and families under reasonable conditions may forgo rofessional or continued education, oting to migrate abroad to high aying unskilled jobs, esecially hen those jobs are aying multiles of their home ay, even for skilled migrants taking unskilled host country jobs. Such an income ga might lead to the rejection of rofessional education and training by individuals in migrants' home countries in exectation of migrating. We kno from observation as ell as from the theoretical literature that such extreme ay gas do exist and can be sustained over the long term as argued by Kravis and Lisey (1983), Bhagati (1984), Panagariya (1988), and Feldman and Gang (1990). Where the existence of high aying lo-skilled jobs abroad reduces educational investment at home, this can give rise to a forsaken schooling tra. Moreover, remaining migrant family members may choose to restructure their houseork because of the migrant's absence. The increase of houseork for children might result in their droing out of school. Parental absence because of migration often means less arental control over children's education, ith erhas sloer rogress in school (Antman, 2012; Elsheniti, 2014). The remaining young adults might anticiate joining their migrant relatives in unskilled ork abroad and choose not to attend school beyond mandatory levels. 2 This is a demonstration effect as the brain-gain literature has argued; only here it orks in the oosite direction hat Stark and Byra (2012) refer to as a back-door brain-drain. Here the demonstration effect results in a loss, leaving the country traed. Since skills and education are major drivers of economic groth and develoment, as a result of foregone schooling the migrant sending country might not rogress further toards having a knoledge-based, advanced economy. Stark and Byra (2012) argue that eventually migrant sending countries suffer from reduced aggregate skill formation, roducing more unskilled and feer skilled laborers. The country might be caught in a lo level equilibrium from hich it is difficult to escae one hich is stable at a loer education level then is good for generating groth. Indeed, it may be less costly for non-rofessionals to accet lo-skilled jobs (Fields, 1974). 2 In related ork, Abdulloev (2013) demonstrates the existence of a discouraged orker effect among relatives of migrants. Page 3 of 34

6 The tye of situation e describe is art of the conventional isdom. With the aim of maximizing the standard of living at home, each individual (informed by family and other constraints) during her or his lifetime has to decide on her or his occuation. Consider a simle orld ith to basic job-tyes: rofessional (skilled, highly educated) or non-rofessional (unskilled or at least not-as-skilled). If the individual chooses a rofessional occuation, his exected income is higher than that of the loer occuation level, but most likely he has a lo emigration robability since ith a rofessional occuation he cannot ork easily in a host country under the same conditions as local rofessionals. What e are saying is that since many host countries establish barriers to foreigners emloyment (ork visas, quotas, local licenses, etc.), it may be difficult for immigrants ho are rofessionally educated to find high quality jobs that ay ell in the host country. Professionals might end u in the same immigrant jobs as non-rofessionals. On the other hand, if an individual decides to forgo studying and take u unskilled emloyment, i.e. ithout acquiring rofessional education, he still has the oortunity to migrate and find a better aid (but still lo skilled) job abroad. Of course, such decisions on acquiring rofessional education in the home country deend on a variety of factors exected earnings in the host and home country, individual status, the robability of finding a job in the to countries, etc. Yet, the incentives for investing in higher level education may become quite lo. In the next section, e rovide an emirical study based on the case of Tajikistan. We chose this former Soviet Reublic because it has high general education comletion rates (i.e. u to the stage of deciding on continuing to rofessional studies or entering the labor force) and significant external migration involving aroximately one-third of its labor force. 3 The main destination of Tajikistan's migrants is Russia as a consequence of the jobs available and the existence of regional free labor movement agreements (Mughal, 2007). There are large age differentials beteen the countries: average monthly ages in Tajikistan ere 8.5 times loer than in Russia (Statistical Committee of CIS, 2011). In section 3 e take u the question of statistical identification. Section 4 discusses several emirical robustness checks. In section 5, e resent a ossible exlanation for our results using a theoretical model of the decision to acquire rofessional education at home and discuss conditions under hich 3 The henomena e are highlighting are not restricted to Tajikistan. There is evidence in the literature of similar declines in educational attainment in Sain during the housing burst, in the choices made by Thai rural-to-urban migrants, among home country households of Albanian emigrants, Turkish migration to Germany, and so on. Page 4 of 34

7 the ossibility of migrating may decrease the robability of individuals choosing such an education. Section 6 concludes. 2. Emirical study: Case of Tajikistan Tajikistan remains the oorest among former Soviet countries ith 31.3% of its oulation living belo the overty line in 2015 (World Bank, 2017) and significant external migration aroximately one-third of the labor force is orking abroad. The main destination of Tajikistan's migrants is Russia (Abdulloev, Estein & Gang, 2015). Remittances and migration are laying an imortant augmenting role in lives of Tajik families: migrants households finance u to half of their consumtion through remittances (World Bank, 2009). Tajikistan as the most externally deendent economy orldide ith remittances comrising 47% of its GDP in At the same time literacy is high in Tajikistan. It inherited a Soviet system of education hich requires all children at age 7 to attend elementary schools, and guarantees their free education, in general basic schools until the age of 16. With enforced free comulsory education, there is no significant effect of migration on children s education; for examle, in households ith a current migrant 8% of children ages 8 to 15 have not been in school in comarison to 6% of children in households ithout a migrant. After comleting their general education at ages 16-17, a young adult can choose either to continue their schooling and acquire rofessional education or enter the labor force. The migration choice is attractive for graduates as ages in Russia are eight times higher than in those in Tajikistan (Statistical Committee of CIS, 2011; for evidence on the rofessional age remium, see Strokova and Ajad, 2017). As a art of the resonse to the recognition of current migration trends and other related social issues in Tajikistan, data ere collected in 2007 for the Tajikistan Living Standards Measurement Survey (TLSS, 2007) highlighting migrants and their families. 4 The survey asks questions on household and individual characteristics, including information on current and returned migrants. 5 Additional surveys asking questions overlaing ith the 2007 survey and 4 Earlier household level data collection efforts and hat they tell us about migration, remittances and education are discussed in Mughal (2007). 5 The survey as conducted in to arts: the first art as in Setember-October 2007, i.e. during the Ramadan month; the second art as conducted after Ramadan, in October- November Hoever, some households ere not visited during the second art: 54 Page 5 of 34

8 samling subsets of the 2007 households ere conducted in 2009 and We discuss the latter to aves later in this aer. We excluded from the samle those resondents ho ere still in school as e ant to distinguish beteen decisions on finishing schooling and migration. 7 Our samle includes individual resondents, orking ages of The total samle size is 16,506 eole, ith 1,536 migrants (9.3%) and 14,970 non-migrants. Migration is almost strictly ork related. Migrants are defined as those individuals ho orked abroad and returned to their homes ithin last 12 months, as ell as those ho ere orking abroad at the time of the survey. Schooling is generally comleted rior to migration. Out of 590 returned migrants only 18 (3%) ere enrolled in the academic year of 2006/2007. Furthermore, e exclude those ho ere in school during the last academic year. This treatment of the samle ill allo us to better to estimate the correlation beteen the schooling and robability of migration regardless of hether the migrant is current or a recent returnee. The survey as erformed during autumn, hich is the season hen migrants generally return back to Tajikistan (due to cold eather conditions in the rimary host, Russia). The samle statistics ith division into migrant and non-migrant subsamles are reorted in the Table 1. Migration in Tajikistan is male dominated; about 93% of migrants are men. In the non-migrant samle 57% are omen. Non-migrants have more children in their households and live in urban areas. They also have comaratively more access to agricultural land. Overall, migrants have more years of education. Years of schooling ere obtained by converting the school degree of resondents to the number of years hich are normally required for such degrees. Levels of education are defined using dummy variables for individually households due to adverse conditions and 100 households could not be found, and 216 households (in the Sughd rovince) ere surveyed ith the combined questionnaire for the to arts. Our samle excludes those resondents ho aeared only during the second Round of the survey. 6 The first to aves of the survey come from the TLSS administered by the World Bank and UNICEF. The third ave of the anel, the Tajikistan Household Panel Survey (THPS 2011) as designed and imlemented by the Institute for the East and Southeast Euroean Studies as a follo-u of the TLSS (Danzer, Dietz and Gatskova, 2013a). 7 We do not create a still in school category as e study the relationshi beteen comleted education and the migration decision. Students ho are still in school ould be fully correlated ith non-migration decision, resulting in multicollinearity. 8 We also erformed the analysis belo restricting the samle to the years old. The results are consistent. Page 6 of 34

9 comleted degrees (see the Aendix for a more details on this assignment). The comarison of education levels shos that eole ith Table 1. Samle Statistics, TLSS 2007 Variables All Migrant Non-Migrant Mean St.Dev. Mean St.Dev. Mean St.Dev. Migrant Male Age Age-squared Years of schooling Years of schooling-squared Education levels: Primary (Grades 1-4) Basic (Grades 1-8(9)) Secondary General (Grades (11)) Secondary Secial Secondary Technical High Comleted formal rofessional education Household's characteristics: No. of children (ages of <15) Lives in the caital Lives in other urban areas Lives in rural areas On land area used for farming Rented land area used for farming On land area rented out (sotka) Monthly er caita consumtion (in thousands somoni) Observations 16,506 1,536 14,970 rimary and basic education mainly remain in Tajikistan. Comaratively more eole ith education from secondary or vocational (secondary secial and technical) schools are migrants; this is because the non-migrant samle is dominated by omen, ho in Tajikistan at this time did not generally obtain a rofessional degree (Abdulloev, Gang and Yun, 2014). There is no difference in higher level education (degrees received from universities) beteen migrants and non-migrants in the samle. 9 The samle statistics suggest an inverted-u relationshi beteen 9 If our samle is divided into male migrant and male non-migrant subsamles, more eole in male non-migrants have education from vocational and tertiary schools. Moreover, educational differences beteen migrants and non-migrants in Table 1 are not significant, as the samle is not restricted to male subsamle. As the migration is male dominated, e control for gender in our regressions hich sho significant difference in decisions on migration and rofessional education. (For samle statistics restricted to the male subsamle, lease see the Aendix). Page 7 of 34

10 the education and migration. Peole at the loer and higher education levels migrate less than those at middle levels (secondary). 10 We no turn to a more detailed analysis of our samle in order to further study the educationmigration relationshi. We estimate a robit model on the individual decision to migrate as an index function of schooling, individual and household characteristics. Other individual characteristics include variables on individual age, age-squared, and gender. Household (not family) characteristics include variables on the number of children in the household ith ages less than 15, hether the household lives in the rural areas (the reference grou is living in urban areas), monthly consumtion deflated due to regional rice differences, as ell as the household's access to and usage of the agricultural land. The deendent variable is hether an individual is a migrant. We estimate to models: one including years of schooling and its square; a second ith categorical educational levels. The coefficient estimates and their marginal effects for both models are reorted in Table 2. Here too e find an inverted-u relationshi beteen the decision to migrate and education. The schooling variables in the Model 1 sho such a relationshi. The decision to migrate is increasing in schooling the coefficient on years of schooling is ositive and statistically significant the coefficient on years of schooling squared is negative and statistically significant. The marginal effect of schooling has a negative imact on the robability of migration. In the second model, instead of variables on years of schooling and its square, e relax the quadratic form and instead include dummy variables on the obtained levels of education. The reference grou in this model is individuals ho have comleted the minimal education level (390 resondents). The coefficients of the dummy variables on education levels are increasing u to the secondary general and secondary secial education, falling afterards. All education level variable coefficients, excet that on rimary education, are ositive and statistically significant at the 1% level. The estimate on the variable rimary education is also ositive but statistically significant at a loer significance level, 10%. The marginal effects of these variables on the robability of migration also reflect the same inverted-u relationshi beteen 10 We do not include indicator variables for oblast (regio. It is common to distinguish the Pamiris, ho have different language; hoever, the same is true of the Uzbeks, Russians and Kyrgyz s. Using a dummy variable for oblast ould not ork as the Badakhshan is not fully inhabited by one language grou, there also different subethnic grous of Pamiris, and the Mugrhab district in Badakhshan is inhabited redominantly by Kyrgyz. Page 8 of 34

11 the migration decision and education. Peole choose to migrate at middle level education, but are less likely to have migrated ith loer or higher levels of education. Table 2: Probit Regression on Migration Decision, TLSS 2007 (Deendent Variable: Migrant=1/0) Variables Estimates Marginal Effects Model 1 Model 2 Model 1 Model 2 Years of schooling (0.0185) (0.0009) Years of schooling-squared (0.0009) Primary (Grades 1-4) (0.2250) (0.0303) Basic (Grades 1-8(9)) (0.1859) (0.0251) Secondary General (Grades 9-10(11)) (0.1818) (0.0245) Secondary Secial (0.1893) (0.0256) Secondary Technical (0.1912) (0.0258) High (0.1874) (0.0253) Other control variables: Age (0.0109) (0.0109) (0.0002) (0.0002) Age-squared (0.0002) (0.0002) Male (0.0429) (0.0429) (0.0058) (0.0058) No. of children (age of <15) (0.0094) (0.0094) (0.0013) (0.0013) Lives in rural area (0.0372) (0.0372) (0.0050) (0.0050) On land used for farming (0.0003) (0.0003) (0.0000) (0.0000) Rented land used for farming (0.0007) (0.0007) (0.0001) (0.0001) On land rented out (0.0029) (0.0029) (0.0004) (0.0004) Monthly er caita consumtion (in thousands somoni) (0.0868) (0.0883) (0.0118) (0.0119) Constant (0.1935) (0.2495) Observations 16,506 16,506 16,506 16,506 Pseudo R Standard errors in arentheses: <.10, <.05, <.01 The robability of migrating is also increasing ith individual s age but at a sloer ace indicating a similar relationshi as e find for education. Peole are likely to migrate in their middle ages, hen they are the most economically active See the Aendix for a chart relating age to the robability of migrating. Page 9 of 34

12 Coefficients on age and age-squared are ositive and negative resectively, and they are statistically significant. Migration in Tajikistan is male-dominated and the majority of migrants are from the rural areas; the coefficients and marginal effects of being a male or living in rural areas is ositive and statistically different from zero. Having more children decreases the robability of migration; the effect of this variable is negative and also statistically significant. While the effects of on land used for farming or rented out does not have a statistically significant imact on the migration decision, renting land for farming decreases this robability. 12 Household consumtion is ositively related to migration, reflecting the imact of remittances on the migrant s home-households consumtion. We also estimate both models for the subsamle of men; as international migration from Tajikistan is male dominated. The estimates and marginal effects of variables for both models are reorted in Table 3. Not surrisingly, the estimated Model 1 ith continuous variables on years of schooling and years of schooling squared have the same signs and significance level as in our revious full samle regression. This imlies the strong inverted-u relationshi beteen education and the decision to migrate. The overall marginal effect of years of schooling on the robability of migration is still statistically different from zero. Also, e see that the education levels suort the inverted-u relationshi beteen education and migration. The marginal effect of education on the robability of migration increases ith education level, reaching the highest imact at the secondary general education level and then decreases. Having only rimary education does not have any statistically significant effect on the robability of migrating as comared to having no education. Other individual and household characteristics variables are similar but the marginal effects are tice as large in absolute size on the robability of migrating. 12 Land onershi is not endogenous ith migration as the land is state roerty and families are given it for certain eriod of time. On land lots are either smaller in size or non-arable. They are mostly received through residential distribution to rural families. Good arable land is rented. Page 10 of 34

13 Table 3: Probit Regression on Migration Decision: Male Subsamle, TLSS 2007 (Deendent Variable: Migrant=1/0) Variables Estimates Marginal Effects Model 1 Model 2 Model 1 Model 2 Years of schooling (0.0207) (0.0017) Years of schooling-squared (0.0010) Education levels: Primary (Grades 1-4) (0.2486) (0.0615) Basic (Grades 1-8(9)) (0.1986) (0.0491) Secondary General (Grades 9-10(11)) (0.1934) (0.0478) Secondary Secial (0.2016) (0.0499) Secondary Technical (0.2026) (0.0501) High (Tertiary) (0.1988) (0.0492) Other control variables: Age (0.0113) (0.0113) (0.0004) (0.0004) Age-squared (0.0002) (0.0002) No. of children (age of <15) (0.0102) (0.0102) (0.0025) (0.0025) Lives in the rural area (0.0408) (0.0407) (0.0101) (0.0101) On land used for farming (0.0004) (0.0004) (0.0001) (0.0001) Rented land used for farming (0.0007) (0.0007) (0.0002) (0.0002) On land rented out (0.0033) (0.0032) (0.0008) (0.0008) Monthly er caita consumtion (in thousands Somoni) (0.1186) (0.1177) (0.0294) (0.0291) Constant (0.2078) (0.2605) Observations 7,887 7,887 7,887 7,887 Pseudo R Standard errors in arentheses: <.10, <.05, < Potential endogeneity and identification An issue ith our regression analysis is the otential endogeneity of the schooling variables: erhas decisions on schooling are taken simultaneously ith the migration decision. In this case, the estimated effect of schooling might be biased. In order to account for the endogeneity issue of the continuous schooling variables, years of schooling and years of schooling squared, e use the instrumental variable robit model, here the structural equation is on the decision to migrate and the reduced form equation is the decision on years of schooling. Page 11 of 34

14 The excluded exlanatory variable in the structural equation is the schooling of household heads both years of schooling and years of schooling squared. We roceeded by excluding household heads from our samle, using their schooling as an instrument for the education decisions of remaining members. 13 The summary statistics of the ne subsamle ithout household heads are reorted in the Table 4. This table shos that the years of schooling and years of schooling squared of household heads do not significantly differ beteen migrant and non-migrant subsamles, i.e. they do not have any redictive oer on the decision to migrate. Such non-variation of these variables makes them relevant instruments in our analysis. 14 Table 4. Summary Statistics: Samle ith Excluded Heads of Households, TLSS 2007 Variables All Migrant Non-Migrant Mean St.Dev. Mean St.Dev. Mean St.Dev. Migrant Age Age-squared Male Years of schooling Years of schooling-squared Household's characteristics: Years of schooling of Household's Head Squared Years of schooling of Household's Head No. of children (ages of <15) Lives in the rural area On land used for farming Rented land used for farming On land rented out Monthly er caita consumtion (in thousands Somoni) Observations 12,543 1,249 11,294 The estimates of the decision to migrate structural equation and the decision about schooling reduced form equation are from an instrumental variable Probit based on Maximum Likelihood Estimation and are reorted in Table The coefficients on years of schooling and years of 13 We use the IV model to control the decision on the rofessional education of migrants by arental education. The children s choice of level of education is strongly correlated ith arental education, hich as received at during the Soviet system. 14 There should not be mean differences beteen decisions on migration and non-migration; arental education should exlain only the decision on education, but not the migration decision. 15 The arameters are estimated using Maximum Likelihood Estimation (IV Probit). A nice feature of this method is that it allos controlling the endogeneity of individual s years of schooling and years of schooling-squared (by adding the residuals to break the correlation beteen the endogenous exlanatory variable and unobservables) in the likelihood function by Page 12 of 34

15 schooling squared of the household heads both strongly redict schooling decisions by other household members. The coefficient on the variable years of schooling of the household head is ositive and statistically significant. The coefficient on the variable years of schooling squared of the household head is negative and statistically different from zero. These results from the reduced form equation estimation indicate that individual schooling has a strong correlation ith the education of the household head. Table 5. IV Probit Regression for Simultaneous Migration and Schooling Decisions, TLSS 2007 Variables Structural Eq: Reduced Form Eq: Marginal Effects Migration=1/0 Years of Schooling of Structural Eq. Years of schooling (0.1315) (0.0028) Years of schooling-squared (0.0065) (0.0003) Age (0.0195) (0.0056) (0.0004) Age-squared (0.0003) ( ) Male (0.1176) (0.0181) (0.0115) No. of children (age of <15) (0.0111) (0.0051) (0.0016) Lives in the rural area (0.0548) (0.0212) (0.0096) On land used for farming (0.0003) (0.0002) ( ) Rented land used for farming (0.0002) (0.0001) On land rented out (0.0022) (0.0020) (0.0004) Monthly er caita consumtion (in thousands Somoni) (0.1150) (0.0580) ( ) Years of schooling of Household's Head ( ) (0.0004) Squared years of schooling of Household's Head Constant (0.2870) (0.1095) Observations 12,543 12,543 12,543 /athrho (0.1800) /lnsigma (0.0160) Wald test of exogeneity (/athrho = 0): chi2(1) = 11.09, Prob > chi2 = Standard errors in arentheses: <.10, <.05, <.01 After controlling for the endogeneity of the decisions about years of schooling, the estimates on individual years of schooling and years of schooling squared in the structural equation on the migration decision still sho the inverted-u relationshi. The coefficient on years of secifying only the reduced form equation for the years of schooling equation (See Wooldridge, J. M. (2010), ). Page 13 of 34

16 schooling is ositive and statistically significant, hile the coefficient on years of schooling squared is negative and also statistically different from zero. The marginal effect of years of schooling is negative and statistically different from zero. We calculate the marginal effects of the years of schooling on the robability of migrating for Model 1, using the hole samle and male subsamle, at seven oints of comleted years of education at each level of education (no education, rimary, basic, secondary general, secondary technical, secondary secial, and tertiary). As the graduate level is the highest level of education and no other level of formal education exists after it, e do not resent the marginal effect for this level. The same exercise is reeated for the IV Probit Regression of the hole samle. The calculations are reorted in Table 6. When e connect the marginal effects at each level of education ith a line as in Grah 1, it shos a clear concave relationshi beteen the education and migration decisions. Both Grah 1 and Table 6 sho that after receiving the secondary education (at 9-11 years of schooling) the small increase in years of schooling moving toards receiving rofessional education reduces the robability of migration. Page 14 of 34

17 Probability of Migration Table 6. Marginal Effects of Years of Schooling on Probability of Migration for Model 1 Probit and IV Probit Regressions, TLSS 2007 Level of Comleted Years of Marginal Effects Education Schooling All Male Subsamle All ith IV No Education (0.0019) (0.0041) (0.0011) Primary (0.0017) (0.0035) (0.0026) Basic (0.0009) (0.0019) (0.0042) Secondary General (0.0010) (0.0019) (0.0047) Secondary Secial (0.0010) (0.0020) (0.0078) Secondary Technical (0.0012) (0.0023) (0.0050) High (Tertiary) (0.0011) (0.0022) (0.0014) Observations 16,506 7,887 12,543 Standard errors in arentheses. <.10, <.05, <.01 Grah 1. Reresentation of Marginal Effects of Years of Schooling on Probability of Migration for Model 1 Probit and IV Probit Regressions, TLSS Years of Schooling All Male Subsamle All ith IV Page 15 of 34

18 4. Emirical Robustness In this section e highlight the most imortant of the many robustness checks e erformed. These are summarized in Tables 7, 8, 9 and 10 and Grah 2. We relicated the same Models 1 and 2 robit regressions using more recent data for Tajikistan: the 2009 Tajikistan Living Standards Measurement Survey (TLSS, 2009) and the 2011 Tajikistan Household Panel Survey (THPS, 2011) as referenced in the introduction. These additional surveys ask questions overlaing ith the 2007 TLSS survey, samling a subset of the same households (Danzer, Dietz and Gatskova, 2013b). We also estimated the robit regressions for the ooled (anel) samle of the three surveys. 16 The additional model 1 estimates of years of schooling and years of schooling-squared sho the strong inverted-u relationshi beteen education and migration decision. In Model 2, here e estimate the imact of each level of education on the robability of migration, the estimation results indicate that comared to eole ithout education, eole ith loer rofessional (vocational) education tend to migrate more than those ho have high (tertiary) education. Since Model 2 results indicate the imact of each educational level in comarison to the reference grou of eole ithout education, this icture does not reflect ho the increase in schooling has an effect on the robability of migration. In order to estimate the correlation of the small change in schooling on the robability of migration, e estimate the marginal effects of years of schooling using Model 1 at each level of education. By connecting each oint of the estimates, e are able to define ho the correlation of years of schooling diminishes ith the robability of migration. This relationshi is shon in Grah 2. Looking at Grah 2 and bringing forard the robit generated 2007 curve from Grah 1, e see the same basic shae in the relationshi beteen years of schooling and the robability of migration for 2007, 2009 and 2011 something of the inverted-u, ith the robability of migration increasing as schooling increases, eaking and then declining ith still more schooling. Moreover, ith each later year e see that the robability of migration is higher for 16 To check on the sensitivity of our results to the age san of the samle, e also erformed the analysis restricting the samle to the years old (not reorted). The results are consistent. Page 16 of 34

19 those less schooled and loer for those staying in school; more eole lacking rofessional education tend to migrate from Tajikistan than eole ith rofessional education. The education-migration relationshi becomes clearer ith each subsequent survey: more eole lacking rofessional education choose to migrate, hile eole ith rofessional education remain at home. Even though the country has good overall school enrollment rates, there is high youth discouragement in finding jobs after school comletion. More and more choose to not go for higher studies, instead migrating abroad. Ajad and others (2014) reort that the quality of education in Tajikistan is becoming an issue, inducing many to forego "costly" loquality rofessional education. We also estimate linear robability models using to stage least squares and three-stage estimation for systems of simultaneous equations ith head of household s schooling and its square as the exclusion restrictions for the regressions. Results are resented in Table 10, hich also shos the inverted-u relationshi beteen education and migration decisions. Comaring coefficient estimates across the biennial samles, e observe large increases in the robit coefficients on education increase by very large magnitudes from.3779 in 2007 and.1495 in 2009 to in 2011 for rimary education, for examle. The marginal effects are also different by large magnitudes. These results are reeated in the other secifications. These results suggest that education variables become stronger redictors of the migration decision after the 2008 orld financial crisis, reflecting structural changes in the migration rocess in Tajikistan (Danzer & Ivaschenko, 2010). Page 17 of 34

20 Variables Table 7: Probit Regression on Migration Decision (Deendent variable: Migrant==1/0), TLSS 2009 and THPS 2011 TLSS 2009 THPS 2011 Estimates Marginal Effects Estimates Marginal Effects Model 1 Model 2 Model 1 Model 2 Model 1 Model 2 Model 1 Model 2 Years of Schooling (0.0389) (0.0016) (0.0376) (0.0018) Years of Schooling-Squared (0.0018) (0.0017) Primary (Grades 1-4) (0.3002) (0.0467) (0.4551) (0.0841) Basic (Grades 1-8(9)) (0.2544) (0.0395) (0.0992) (0.0260) Secondary General (Grades (11)) (0.2465) (0.0383) (0.0727) (0.0227) Secondary Secial (0.2641) (0.041) (0.1058) (0.0267) Secondary Technical (0.2637) (0.0409) (0.1147) (0.0275) High (Tertiary) (0.2593) (0.0403) (0.0954) (0.0257) Other control variables: Age (0.0173) (0.0175) (0.0004) (0.0004) (0.0161) (0.0160) (0.0004) (0.0004) Age-squared (0.0002) (0.0002) (0.0002) (0.0002) Male (0.0661) (0.0662) (0.0098) (0.0099) (0.0576) (0.0579) (0.0090) (0.0092) No. of children (age of <15) (0.0149) (0.0149) (0.0023) (0.0023) (0.0119) (0.0120) (0.0021) (0.0022) Lives in the rural area (0.0606) (0.0607) (0.0094) (0.0094) (0.0550) (0.0557) (0.0099) (0.0100) Monthly er caita consumtion (in thousands of Somoni) (0.1602) (0.163) (0.0248) (0.0252) (0.0219) (0.0212) (0.0040) (0.0038) Constant (0.3525) (0.3766) (0.3515) (0.2878) Observations 5,647 5,647 5,647 5,647 5,804 5,804 5,804 5,804 Pseudo R Standard errors in arentheses. <.10, <.05, <.01 Page 18 of 34

21 Table 8: Pooled Probit Regression on Migration Decision (Deendent variable: Migrant==1/0), TLSS 2007, TLSS 2009 and THPS All samle: Estimates Male Samle: Estimates Model 1 Model 2 Model 1 Model 2 Years of Schooling (0.0155) (0.0174) Years of Schooling-Squared (0.0007) (0.0008) Education levels: Primary (Grades 1-4) (0.1774) (0.1979) Basic (Grades 1-8(9)) (0.1475) (0.1594) Secondary General (Grades 9-10(11)) (0.1446) (0.1556) Secondary Secial (0.1502) (0.1617) Secondary Technical (0.1512) (0.1621) High (Tertiary) (0.1486) (0.1595) Other control variables: Age (0.0080) (0.0080) (0.0083) (0.0083) Age-squared (0.0001) (0.0001) (0.0001) (0.0001) Male (0.0306) (0.0305) No. of children (age of <15) (0.0066) (0.0065) (0.0072) (0.0071) Lives in the rural area (0.0274) (0.0273) (0.0300) (0.0299) Monthly er caita consumtion (in thousands Somoni) (0.0203) (0.0212) (0.0210) (0.0222) Year of (0.0294) (0.0294) (0.0326) (0.0326) Year of Constant (0.1946) (0.1533) (0.2046) (0.1634) Observations 27,957 27,957 13,426 13,426 Pseudo R Standard errors in arentheses: <.10, <.05, <.01 Page 19 of 34

22 Probability of Migration Table 9. Marginal Effects of Years of Schooling on Probability of Migration for Model 1 Probit Regression, TLSS 2009 and THPS 2011 Level of Comleted Education Years of Schooling TLSS 2009 THPS 2011 Pooled Samle No Education (0.0010) (0.0022) (0.0010) Primary (0.0025) (0.0023) (0.0013) Basic (0.0019) (0.0023) (0.0008) Secondary General (0.0018) (0.0021) (0.0008) Secondary Secial (0.0019) (0.0022) (0.0008) Secondary Technical (0.0024) (0.0023) (0.0010) High (Tertiary) (0.0025) (0.0020) (0.0009) Observations 5,647 5,804 27,957 Standard errors in arentheses. <.10, <.05, <.01 Grah 2. Reresentation of Marginal Effects of Years of Schooling on Probability of Migration for Model 1 Probit Regressions, TLSS 2009, THPS 2011 and Pooled Samle Years of Schooling TLSS 2007 TLSS 2009 TPS 2011 Pooled Samle Page 20 of 34

23 Table 10. Linear Probability Regressions for Simultaneous Migration and Schooling Decisions, TLSS 2007 Linear robability regression: 2SLS 3SLS First stage: Years First stage: Years of Second stage: (1) (2) Variables Of schooling schooling-squared Migrant Migrant Years of schooling Years of schooling (0.0273) (0.0268) Years of schooling-squared ( ) ( ) Migrant (42.14) Other control variables: Age (0.0127) (0.235) ( ) ( ) (0.447) Age-squared ( ) ( ) ( ) ( ) Male (0.0513) (0.953) ( ) ( ) (9.946) No. of children (age of <15) (0.0134) (0.248) ( ) ( ) (0.429) On land used for farming e ( ) ( ) (5.40e-05) (0.0110) Rented land used for farming ( ) (0.0119) (0.0166) On land rented out ( ) (0.126) ( ) ( ) (0.146) Monthly er caita consumtion (in thousands Somoni) (0.192) (3.575) (0.0229) (0.0210) (5.252) Lives in the rural area (0.0554) (1.029) ( ) ( ) (1.455) Years of schooling of Household's Head (0.0208) (0.386) (0.570) Squared years of schooling of Household's Head ( ) (0.0202) (0.0261) Constant (0.245) (4.549) (0.109) (0.107) (9.014) Observations 12,543 12,543 12,543 12,543 12,543 R-squared IV F-stat Durbin val Standard errors in arentheses. <0.01, <0.05, <0.1 Page 21 of 34

24 5. A Possible exlanation - A Story from a Theoretical Model Our emirical ork above shos the form of the link beteen education and migration for a high emigration country. Here e suly a background story, in the form of economic theory and educational choices, to demonstrate our thinking that the ability to migrate and obtain high ages ould/may decrease the desirability of higher schooling. This then, as almost all of the literature tells us, is detrimental to economic groth and elfare. We consider a oulation of orkers hose utility is searable and linear in rivate consumtion that is rovided by exenditure of a age,, and in occuational status, s, W(, s) = + s. (1) The orker s occuation determines status. To simlify e assume orkers are either emloyed in a rofessional occuation or a non-rofessional occuation. We think of status, s, as artly a shorthand for educational achievement. A non-rofessional orker ill have a status level of s=0 hile a rofessional orker ill have a status of s>0. All orkers are risk neutral in income and risk averse to their status. Workers maximize resent discounted utility, ith a rate of time reference r>0. The model is set in continuous time. The only choice that a orker makes is selection of educational tye: one that leads to either a rofessional occuation or one that leads to a non-rofessional occuation. We assume that all orkers have the same abilities; thus, if an individual chooses to study the rofessional track the individual ill succeed in obtaining a rofessional job. In the home market a rofessional orker ill earn and an non-rofessional orker ill earn n.. With robability q() a rofessional orker ill migrate and find a job hile ith robability q( a non-rofessional orker ill migrate and find a job. V() and V( are the exected lifetime utilities of rofessional and non-rofessional orkers, resectively 17. U() and U( are the exected lifetime utilities of rofessional and nonrofessional orkers, resectively, ho have migrated. For a rofessional orker, ( ) = + s + q ) ( U ( ) V ( ) ) rv (, (2) 17 In a similar ay to Estein and Hillman (2003). Page 22 of 34

25 and for a non-rofessional orker, ( + q( ( U( V ( ) rv = n. (3) From (2) and (3), e have, ( ) U ( ) ( ) + s + q V ( ) = and r + q ( U( q( n + q V ( =. (4) r + The age at hich a orker is indifferent beteen being a rofessional or a non-rofessional ill satisfy ( ) V( hile if ( ) V( V =. If V( ) V(, the individual becomes a non-rofessional orker V the individual becomes a rofessional orker. Denote by the age rofessional orkers are aid at hich they are indifferent beteen becoming a rofessional and a non-rofessional. From (4) e comute ( n + q( U( )( r + q( ) ) r + q(, + ( s q( ) U( ) ) =. (5) If ages in the rofessional occuation are less than, this individual ill decide not to obtain rofessional qualifications. On the other hand, if the ages the rofessional orker earns are at least, then this individual ill decide to become a rofessional orker. No let us consider ho changes as arameters change. Increasing the reutation and status (s) that a rofessional obtains from his rofession decreases the age that makes the individual indifferent beteen occuations. Thus as the status (s) of rofessionals increase, more individuals become rofessionals, 0. What this says that as the status a orker s receives from becoming a rofessional increases, the age that ill make him indifferent beteen being and not being a rofessional decreases, thus increasing the number choosing a rofessional occuation and the extra schooling this choice entails. If individual references for the resent (r) increase, e have r = ( ( )) ( q( q( ) ) n + q( U n ( r + q( ) 2. The sign of this exression deends on the difference Page 23 of 34

26 beteen the robabilities of emigration for the to occuations. If the robability of a rofessional orker emigrating is loer than the robability of a non-rofessional orker emigrating, ( q ( q( ) ) 0, then increased reference for the resent increases the age that makes the individual indifferent. The main idea is that as time reference increases, the eight on the future decreases and eole care more about the resent, thus if the robability of migrating for a rofessional orker is loer than for a non-rofessional orker, the ages needed for a orker i to choose the rofessional occuation increase and the age has to go u to make him ant to migrate. Emigration otions,, and choice of rofession Let us no consider ho different emigration ossibilities affect. As the exected lifetime utility of rofessional orkers ho emigrate, U(), increases, the age that makes the individual indifferent decreases, U ( ) 0, as he can gain more by emigrating. And as the exected lifetime utility of non-rofessional orkers ho emigrate, U(, increases, the age that makes the individual indifferent increases, U ( 0, since the oortunity cost has increased. This is a comarative static calculation. The effect of a change in the migrant's utility on deends on the utility the migrant obtains after migration. If a rofessional orker obtains a high utility after migration, then he ill be illing to become a rofessional at a loer age (loer ); hoever, if there are more benefits for a nonrofessional orker in the host country, then the orker needs a higher age to become a rofessional. Thus, it all deends on the oortunities the orker may have in different locations. For locations that individuals can migrate to that have more oortunities for rofessional orkers (relative to nonrofessional orkers) the value of decreases, hile for locations that a nonrofessional orker has more oortunities (relative to a rofessional orker) the value of increases. In other ords, one can think of it as an otion to migrate to different laces ith different exected income. Note that if then individuals ill choose a rofessional occuation. With robability q() the individual after becoming a rofessional orker ill emigrate. Hoever, ith Page 24 of 34

27 robability 1- q(), the individual ill not emigrate and ill stay at in the home country. If earnings in the host country are such that this age is sufficiently large, then individuals are choosing a rofessional occuation in order to emigrate and earn high ages in the host country. Hoever, only a roortion q() actually emigrate hile a roortion 1- q() do not emigrate. In the literature this is called a brain-gain since some individuals ho decided to become rofessionals in order to emigrate in the end stayed in their home country adding to its human caital. Hence, the home country benefits from the ossibility of emigration. Introducing return migration into the model To better understand the results, consider the extreme case here the robability of migrating is lo for the rofessional orker hile it is higher for the non-rofessional orker. For emhasis, and only for emhasis reasons, assume that rofessionals have a zero robability of emigrating q() = 0. Changing this to a ositive robability ill not change the main results resented belo. Exlicitly rite the lifetime utility of a migrant taking into consideration the robability of return migration. With robability k( a non-rofessional migrant ill return home. This can haen, for examle, as a result of being illegal and thus arehended and deorted, q1, or the immigrant simly decides to return home ith robability q2. An immigrant s utility is indeendent of the identity of his or her revious emloyer. Hence, ( k ( U ( V ( ) ru = f, (6) here, f is the immigrant s monthly income in the host country, U( is the exected utility of an emloyed non-rofessional orker ho migrated, and V( is that orker s exected lifetime utility. From (6) e obtain that U ( ( f + V =. (7) k + r Rerite the lifetime utility of an individual taking into consideration comonents determining the lifetime utility of an immigrant. Substituting (7) into (4), e obtain Page 25 of 34

28 ( f + V n + q( k r V ( + =. (8) r + q( Thus, n ( k + r) + q( f ( r + q( )( k + r) q( V ( =. (9) Let us use this to udate the age that makes the individual indifferent beteen choosing to be a rofessional and a nonrofessional orker. Comaring (9) ith s V ( ) = + (the r exected discounted utility of a rofessional orker ith a robability of migrating equaling zero, q()=0), e see that the age of a rofessional orker that ill make the individual indifferent beteen becoming a rofessional orker or a non-rofessional orker ill satisfy V ( ) = V(. If V( ) V( then the individual becomes a non-rofessional orker; if V( ) V( the individual becomes a rofessional orker. Denote by the age a rofessional orker must earn to make him indifferent beteen becoming a rofessional orker and a non-rofessional orker. From (4) e comute, r + s = n ( k + r) + q( f ( r + q( )( k + r) q(. (10) Thus, hen accounting for the ossibility of return migration, the age that a rofessional has to earn in the host country to make an individual indifferent beteen this occuation and being a non-rofessional, must satisfy = r n ( k + r) + q( ( r + q( )( k + r) ) f q( s. (11) This age determines hether an individual ill be a rofessional or non-rofessional orker. Consider ho it changes ith changes in the arameters. Increasing the reutation and status (s) a rofessional orker obtains from his rofession ill decrease the age that makes the individual indifferent beteen the rofessions. Thus as status, s, increases more individuals Page 26 of 34

29 ill stay in school and become rofessional orkers, 0. As the income of a nonrofessional orker in the home country or host country increases, n or f, the age s that makes the individual indifferent beteen the to otions increases 0 and 0 n f. As the robability of migration increases, q(, the age that makes the individual indifferent, ill increase: = r q( ( f + n n ( k + r) ) (( r + q( )( k + r) q( ) 2 0 since r + k 2 and f n. Thus if monthly ages in the host country are higher than those in the home country, f n, then increasing the robability of migration ill increase the chances an individual ill choose a non-rofessional occuation. The story e have just told examines the henomenon of forsaken schooling as a result of oortunities abroad. While high-skilled migration is orrisome, many international migrants accet lo-skilled ositions in host countries. Their illingness to do so arises from very large host-home earnings differentials. If the ages of rofessional orkers are not sufficiently high, individuals ill decide not to become rofessionals since the chances of migrating and earning a higher discounted utility ill be better for non-rofessionals. At home this can lead to reduced educational investment as eole forgo additional schooling because of oortunities to migrate to high aying lo-skilled jobs the market is discouraging eole from investing in education. This suggests there might be time-inconsistencies beteen short-run economic gains from migration and negative long-term effects from missing human-caital investment. Our modeling allos us to establish the circumstances under hich this tye of forsaken schooling ill occur and the trade-offs that olicymaker s need to consider (Estein, 2013 and Estein and Gang, 2010). 6. Conclusion This aer shos ho oortunities for international migration may affect the education decisions of otential migrants. Tajikistan is an excellent choice for a case study of this relationshi because of the high rate of labor migration from Tajikistan (mostly to Russia) and the imortance of remittances for household income and consumtion. Tajikistan also has excellent household level data ith hich to look at the relationshi beteen migration and Page 27 of 34

30 education, and the last three TLSS surveys form a anel that can be used to observe changes in education and migration in the same households over time. The toic is also imortant for education and labor olicy formulation. Underinvestment in education may have short run benefits to households, but, in the long run, economic groth and develoment can be negatively imacted by the lack of skill in society. Existing age differences in migrant host and home countries might induce eole in the home country to forgo rofessional education, oting to migrate abroad for high aying unskilled ork. We sho ho an individual s choice of hether to ursue rofessional education might be affected by the oortunity to migrate. Oting for higher education rovides a higher exected income at home than for unskilled labor, and the otential migrant ill face a loer emigration robability. Hoever, ith large enough international age differentials even rofessionals ill migrate, and since it is difficult for rofessionals to find a high quality and ell aid job in the host country, many end-u taking unskilled jobs hich still dominate their home country rofessional incomes. The next or near-next grou making the decision to continue their schooling (once it becomes otional) or not, may forgo rofessional education as their exected earnings in the home country are lo relative to otential migrant destinations. The orker takes the oortunity to migrate and find a better-aid unskilled job abroad. We offer a ossible exlanation for these results in a theoretical model that enables us to understand better our emirical results. Our theory exlains the forsaken schooling henomenon as a result of lo-skilled and skilled orkers acceting lo-skilled ositions in host countries, hich leads to the forgoing of rofessional schooling in the home country. We exect, therefore, eole ho decide to migrate abroad to have either loer years of schooling, or generally have not comleted rofessional schools (technical-vocational or tertiary). Of course, our theoretical model is one out of many ossible exlanations. Using the case of literate Tajikistan, the most remittance deendent country in the orld, e sho that the decision to migrate is a concave function of education. The robability of migrating is increasing ith non-rofessional (rimary, basic and general secondary) education, but shrinks for those ith rofessional education. This relationshi remains robust even after accounting for the endogeneity of years of schooling. Such results suggest that eole in Tajikistan ill choose to forgo the rofessional education in favor of migration. This can give rise to a foregone schooling tra, here the existence of high aying lo-skilled jobs abroad reduces educational investment. Page 28 of 34

31 The education-migration relationshi becomes clearer over subsequent survey years, imlying that more eole ithout rofessional education choose to migrate, hile eole ith rofessional education remained in their home country. Even though the country has good overall school enrollment rates, the youth are highly discouraged by their failure in finding jobs after comleting school, and therefore choose to not go for higher studies, but migrate abroad. Page 29 of 34

32 References Abdulloev, I. (2013) Imact of Migration on Job Satisfaction, Professional Education and the Informal Sector. PhD Dissertation, Rutgers University, Ne Brunsick, NJ. Available at: htts://rucore.libraries.rutgers.edu/rutgers-lib/41526/ Abdulloev, I., Estein, G.S., & Gang, I.N. (2015). Ethnic goods and immigrant assimilation. In P. Nijkam, J. Poot, & J. Bakens (Eds.), Economics on Cultural Diversity ( ) Cheltenham, UK: Edard Elgar Pub. Abdulloev, I., Gang, I. N., & Yun, M. S. (2014). Migration, education and the gender ga in labour force articiation. Euroean Journal of Develoment Research, 26(4), Ajad, M. I., Hut, S., Abdulloev, I., Audy, R.. de Laat, J., Kataoka, S., Larrison, J., Nikoloski, Z., and Torracchi, F. (2014). The Skills Road: Skills for Emloyability in Tajikistan. World Bank, Washington, DC. Acosta, P. (2006). Labor suly, school attendance, and remittances from international migration: the case of El Salvador. World Bank Policy Research Working Paer, 3903, World Bank. Amuedo-Dorantes, C., & Pozo, S. (2010). Accounting for remittance and migration effects on children s schooling. World Develoment, 38(12), Antman, F. M. (2012). Gender, educational attainment, and the imact of arental migration on children left behind. Journal of Poulation Economics, 25(4), Beine, M., Docquier, F., & Raoort, H. (2008). Brain drain and human caital formation in develoing countries: Winners and losers. The Economic Journal, 118(528), Beine, M., Docquier, F., & Raoort, H. (2001). Brain drain and economic groth: theory and evidence. Journal of Develoment Economics, 64(1), Bhagati, J. N. (1984). Why are services cheaer in the oor countries?. The Economic Journal, 94(374), Bhagati, J., & Hamada, K. (1974). The brain drain, international integration of markets for rofessionals and unemloyment: a theoretical analysis. Journal of Develoment Economics, 1(1), Calero, C., Bedi, A. S., & Sarro, R. (2009). Remittances, liquidity constraints and human caital investments in Ecuador. World Develoment, 37(6), Co, C.Y., Gang, I. N., & Yun, M. S. (2000). Returns to returning. Journal of Poulation Economics, 13(1), Dai, T., Liu, X., & Xie, B. (2015). Brain drain reversal and return subsidy. Journal of Comarative Economics, 43(2), Danzer, A. M., Dietz, B., & Gatskova, K. (2013a). Tajikistan Household Panel Survey: Migration, Remittances and the Labor Market. Survey reort, IOS Regensburg. Danzer, A. M., Dietz, B., & Gatskova, K. (2013b). Migration and remittances in Tajikistan: Survey technical reort (No. 327). IOS Working Paers. Danzer, A.M., & Ivaschenko, O. (2010). Migration atterns in a remittances deendent economy: Evidence from Tajikistan during the global financial crisis. Migration Letters, 7(2), Dimova, R. (2017). Migrant Remittances and Beyond: The develoment imlications of human caital mobility and accomanying financial tools in G. Giorgioni (ed), Develoment Finance, Palgrave Studies in Imact Finance, Page 30 of 34

33 Dimova, R., Estein, G. S., & Gang, I. N. (2015). Migration, Transfers and Child Labor. Revie of Develoment Economics, 19(3), Duryea, S., Cox, A., & Ureta, M. (2003). Adolescents and human caital formation. In Duryea, S., Cox, A., & Ureta, M. (eds), Critical decisions at a critical age: Adolescents and young adults in Latin America, Inter-American Develoment Bank, Edards, A. C., & Ureta, M. (2003). International migration, remittances, and schooling: evidence from El Salvador. Journal of Develoment Economics, 72(2), Elsheniti, O. (2014). Intergenerational mobility, income inequality and children's human caital investment. PhD Dissertation, Rutgers University, Ne Brunsick, NJ. Estein, G. S., & Hillman, A. L. (2003). Unemloyed immigrants and voter sentiment in the elfare state. Journal of Public Economics, 87(7), Estein, G. S. and Gang, I. N. (2010). A olitical economy of the immigrant assimilation: internal dynamics in G. S. Estein, and I. N. Gang, (eds.), Migration and Culture: Frontiers of Economics and Globalization, volume 8, Emerald Grou Publishing Limited. Estein, G. S. (2013). Issues on the olitical economy of migration in A. Constant and K. F. Zimmermann (eds.) International Handbook on the Economics of Migration ( ) Cheltenham, UK: Edard Elgar Pub. Feldman, D. H., & Gang, I. N. (1990). Financial develoment and the rice of services. Economic Develoment and Cultural Change, 38(2), Fields, G. S. (1974). The rivate demand for education in relation to labour market conditions in less-develoed countries. The Economic Journal, 84(336), Grubel, H. B., & Scott, A. D. (1966). The international flo of human caital. The American Economic Revie, 56(1/2), Kravis, I. B., & Lisey, R. E. (1983). Toard an exlanation of national rice levels. Princeton Studies in International Finance, no. 52. Princeton, N.J., November Mountford, A. (1997). Can a brain drain be good for groth in the source economy?. Journal of Develoment Economics, 53(2), Mughal, A. (2007). Migration, remittances, and living standards in Tajikistan. International Organization for Migration, Dushanbe. Özden, Ç. G. (2006). Educated migrants: is there a brain aste? In Ç. G. Özden, M. Schiff (Eds.), International migration, remittances, and the brain drain ( ). Washington, DC: World Bank. Panagariya, A. (1988). A theoretical exlanation of some stylized facts of economic groth. The Quarterly Journal of Economics, Piracha, M., Randazzo, T., & Vadean, F. (2013). Remittances and occuational outcomes of the household members left-behind. IZA Discussion Paer No Available at SSRN: htt://ssrn.com/abstract= Stark, O., & Wang, Y. (2002). Inducing human caital formation: migration as a substitute for subsidies. Journal of Public Economics, 86(1), Stark, O., & Byra, L. (2012). A back-door brain drain. Economics Letters, 116(3), Stark, O., Helmenstein, C., & Prskaetz, A. (1997). A brain gain ith a brain drain. Economics Letters, 55(2), Stark, O., Helmenstein, C., & Prskaetz, A. (1998). Human caital deletion, human caital formation, and migration: a blessing or a curse?. Economics Letters, 60(3), Page 31 of 34

34 Statistical Committee of CIS (2011) Average Monthly Nominal Wage in the CIS Countries, in national currency. Available: htt://.cisstat.com/index.html [Accessed 17 Setember 2011/23 December 2015]. Strokova, V., & Ajad, M. I. (2017). Tajikistan Jobs Diagnostic: Strategic Frameork for Jobs. Job series; no. 1. Washington, D.C.: World Bank Grou. htt://documents.orldbank.org/curated/en/ /tajikistan-jobs- Diagnostic-Strategic-Frameork-for-Jobs Weiss, Y., Sauer, R. M., & Gotlibovski, M. (2003). Immigration, search, and loss of skill. Journal of Labor Economics, 21(3), Wooldridge, J. M. (2010). Econometric analysis of cross section and anel data. MIT ress. World Bank (2009) Reublic of Tajikistan Poverty Assessment: Reort No TJ. Human Develoment Sector Unit, Central Asia Country Unit, Euroe and Central Asia Region. Available: htts://oenknoledge.orldbank.org/bitstream/handle/10986/3159/513410esw0p Official0Use0Only1.txt?sequence=2 [Accessed 23 December 2015]. World Bank (2015). World Data Bank. Available: htts://data.orldbank.org/country/tajikistan [Accessed 22 December 2017]. Yang, D. (2005). International migration, human caital, and entrereneurshi: Evidence from Philiine migrants' exchange rate shocks. World Bank Policy Research Working Paer The World Bank. Page 32 of 34

35 Aendix Schooling, Age and Degrees The folloing chart lays out the structure of Tajikistan s educational system, the translation of degrees into years of schooling, and the normal corresonding students ages Universities, Institutions and the Conservatorium 21 5 Technical 20 4 Lyceums & schools & Colleges 19 3 Vocational schools Lyceums 18 2 &Vocational 17 Comlete Secondary Education 1 schools 16 Years Comulsory - Basic Secondary Education - 9 years - until the age of 16 Age In this aer, rofessional education starts at years of schooling category Professional lyceums and Vocational schools corresonding to age 16 or 9 years of comulsory basic education. These are the loer band for the rofessional education categories. With rofessional education from lyceums and vocational schools, eole ork at lo rofessional occuations. For high education at least 16 years of schooling are needed: 11 years of general + 5 tertiary. Male Samle Statistics, TLSS 2007 Variables All Migrant Non-Migrant Mean St.Dev. Mean St.Dev. Mean St.Dev. Migrant Age Age-squared Years of schooling Years of schooling-squared Education levels: Primary (Grades 1-4) Basic (Grades 1-8(9)) Secondary General (Grades (11)) Secondary Secial Secondary Technical High Comleted formal rofessional education Household's characteristics: No. of children (ages of <15) Lives in the rural area On land used for farming Rented land used for farming On land rented out Monthly er caita consumtion (in thousands Somoni) Observations Page 33 of 34

36 Probability of Migrating by Age Page 34 of 34