Impacts of migration and migrant’s gender on children’s school enrollment and child work in viet nam

From the first stage of migrant indicator, it indicates that instrument historical migration network and number of male adults will impact on migrant indicator significantly and these in

i

UNIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES

HO CHIMINH CITY THE HAGUE

VIETNAM THE NETHERLANDS

VIETNAM - NETHERLANDS

PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

IMPACTS OF MIGRATION AND MIGRANT’S GENDER

ON CHILDREN’S SCHOOL ENROLLMENT AND CHILD

WORK IN VIET NAM

BY

VÕ THỊ THU HOÀI

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

HO CHI MINH CITY, APRIL 2014

ii

U NIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES

HO CHIMINHCITY THE HAGUE

VIETNAM THE NETHERLANDS

VIETNAM - NETHERLANDS

PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

IMPACTS OF MIGRATION AND MIGRANT’S GENDER

ON CHILDREN’S SCHOOL ENROLLMENT AND CHILD

WORK IN VIET NAM

A thesis submitted in partial fulfilment of the requirements for the degree of

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

Secondly, I would like to thank Dr Truong Dang Thuy because of his dedicated support in research method during the time of this thesis All of his help strongly to find the solution and improves this paper

Besides, my sincere thanks also go to Dr Nguyen Trong Hoai, Dr Pham Khanh Nam who supervised and motivated the Class MDE 17 to finish the course on time

It is grateful to thank my classmates of MDE17 MDE18, VNP staffs for stimulating discussions, for the fun time we had together

Last and not the least, I would like to express my grateful thank to my family who are always beside me, give me the birth and support me throughout my life

April, 2014

VO THI THU HOAI Email: hoai.vtt@vnp.edu.vn

v

ABSTRACT

Accompany with development trend in over the world, migration flow plays important role in Vietnamese economy It contributes to the significant income and raises the living standard for household in the country, specially, in rural areas This paper tries to measure the impact of migration on children’s schooling enrollment and child work, addition to, determine how migrant’s gender matter in this impact The context is applied

in rural areas in Vietnam with the dataset of VHLSS 2010 by Instrument variable method

to deal with the problem of endogeneity of migration From the first stage of migrant indicator, it indicates that instrument historical migration network and number of male adults will impact on migrant indicator significantly and these instruments are the strong instruments The results show that the presence of migrant in the household will make the children take part in school more, at the same time make children work less Besides, not like others researches, gender of migrant and time of migrant using in household don’t have meaning with children’s welfare

Keywords: migration, migrant’s gender, children’s school enrollment, child work

vi

TABLE OF CONTENT

ABSTRACT v

CHAPTER 1: INTRODUCTION 1

1.1 Problem statement 1

1.2 Research Objectives 3

1.3 Research question 3

1.4 Research scope 3

1.5 Structure of the research 3

CHAPTER 2: LITERATURE REVIEW 5

2.1 Definition of key concept 5

2.2 Theoretical literature 5

2.3 Empirical literature 12

2.4 Conceptual framework 16

CHAPTER 3: METHODOLOGY 18

3.1 Endogeneity problem 18

3.2 Endogeneity of migration 18

3.3 Estimated equation 20

3.3.1 Validity of Instrument variable: 20

3.3.2 IVs methods 20

3.3.3 Estimated equation: 24

3.3.4 Method to run IVs regression 26

3.4 Data 28

3.4.1 Source of data 28

3.4.2 Variables description and measurement: 29

CHAPTER 4: OVERVIEW OF MIGRATION IN CASE OF VIETNAM 36

4.1 Socio-economic setting and migration in Vietnam 36

4.1.1 Migration aboard 36

4.1.2 Internal migration 37

4.2 Characteristics of migrant 39

vii

4.2.1 International migrant 39

4.2.2 Internal migrant 44

CHAPTER 5: EMPERICAL ANALYSIS 46

5.1 Descriptions of variables 46

5.2 Estimation results 51

5.3 Interpretation of results 57

CHAPTER 6: CONCLUSIONS AND RECOMMENDATIONS 63

6.1 Conclusions 63

6.2 Recommendations: 64

6.3 Limitations 65

REFERENCE 67

APPENDIX 70

viii

LIST OF FIGURES

Figure 2.1: Conceptual framework about impact of migration and migrant’s gender on

children’s welfare 1717 Figure 3.1: Histogram of expenditure per capita a year 31 Figure 3.2: Histogram of natural logarithm of expenditure per capita a year 31 Figure 4.1: International migration trend from 2000 to 2010 (Department of oversea

database) 40 Figure 4.2: Number of male and female international migrant from 2006-2010 (IOM) 41 Figure 4.3: Main destinations of international migrant from 2000-2010 (IOM) 42 Figure 4.4: Structure of international migration labor of Viet Nam from 2006-2010 (IOM, 2011) 43

ix

LIST OF TABLES

Table 3.1: Description and measurement of variables 32 Table 5.1: T-test between household with non-migrant and household with migrant 48 Table 5.2: T-test between household with male migrant and household with female migrant 50 Table 5.3: Factors affecting migration indicator in the household (results of first stage

regression of Instrumental variables) 53 Table 5.4: Factors affecting children’s school enrollment and child work (results of second stage regression of Instrumental variables) 56

x

ABBREVIATIONS

BLUE: Best linear unbiased estimates

DID: Difference in difference

GSO: General statistic office

IOM: International Organization for Migration

IVs: Instrumental variables

OLS: Ordinary least square

PSM: Propensity score matching methods

VHLSS: Vietnam Household Living Standard Survey

UN: United Nations

UNDP: United Nations Development Programme

US: United Stages

From economic renovation in 1986, Viet Nam had many remarks in development process Among them, shifting from centrally-planned economy to a socialist-oriented market economy for industrialization, modernization was a long step in policies and economic aspect Beside the diversification and multilateral development of open-door economic, more and more Vietnamese aboard to live and work External migration contributes to economy and raises the living standard of remaining people In 2012, Viet Nam was rated in seventh among the biggest receiving remittances countries ($ 10 billion) with about 4 million oversea Vietnamese From 1991, it was accounted about 60-70% remittances comparing to foreign investment Remittance not only plays an important role in Vietnam’s economy, but also affects to the demographic of Vietnamese people According to United Nations Development Programme (2009), estimated amount

2

of money which migrant send back to their family reaches 5.5 billion dollars in 2007 In parallel with that, internal migration from rural to urban has become a general upward trend in Viet Nam a few last year Not only male, the share of female migrants has strongly dominated and played an important role in migrant worker (a half of internal migration in 2009) Most migrants have close relationship with their origin families through sending remittances and affecting demographic characteristics of members in the household It is so useful for examining the impacts of migration on children’s welfare In reality, education investment for children occupies a big part of expenditure amount in a household So, the matter is whether migration contributes to school enrollment of children or not and how children in migrant household will differ with no migrant household’s children From that, need deeper research about whether gender of migrant and time of migrant living in household can impact directly to school enrollment Some suggestions can be concluded in reality which need to be confirmed are: (1) Migrant help children have more school enrollment; (2) if household has female migrant, the level of school enrollment of children is better than household having male migrant Two issues, school enrollment and child work, of course, are interconnected closely When migration affects to children’s school enrollment, it will affect to child work also Besides that, there were many reports and papers such as Cuong (2008), Pfau and Long (2006) research about migration and remittance before Meanwhile impact of migration and migrant’s gender on children’s school enrollment and child work in Vietnam has not been studied and need more papers having the deep view about this aspect

By applying data of Vietnam household living standard survey (VHLSS) 2010, this paper is expected to use a suitable regression to investigate the impacts of migration and migrant’s gender on children’s school enrollment and child work in Viet Nam for that time clearly After that, a main key factor of migrant which wasn’t recommended in the similar previous research, that is time of migrant spending in household will be used to

3

find out how migrant’s monitor effect children’s welfare Finally, it hopes that the results

of research will be helpful for migration policy

1.2 Research Objectives

The aiming objectives of this research are to meet the following objectives:

a) To examine the impacts of migration and migrant’s gender on children’s school enrollment and child work

b) To make recommendations for government about the policies and programs which tend to promote the effectiveness of migration in children development

as well as human capital investment

1.3 Research question

The research will try to address some questions like that:

a) How do migration and migrant’s gender significantly affect children’s school enrollment?

b) How do migration and migrant’s gender significantly affect child work?

1.5 Structure of the research

This research will be classified into 6 main parts The first part introduces overview about the research, raising the problem and objectives as well as scope of the research

4

The second part is the literature is the solid basis theory and previous empirical experience before relating to research The third part shows methodology, estimated equation in the research It also talks about the data which used for methodology The next part is to know how social-economic characteristic of Vietnam impact on migration and overview of migration case in Vietnam The fifth part show the result of model after running the regression, and explain the result appear in the model The final part is conclusion and recommendation, limitation need to take experience in the research

5

CHAPTER 2: LITERATURE REVIEW

2.1 Definition of key concept

Children: United Nations Convention on the Rights of the Child considers children as

a human being under 18 years old Even in some definitions, child includes the fetus and the unborn It is important to note that in most paper, the author will define children by their own purpose by disaggregating children group into specified group For the purpose

of this research relating to school and labor supply of child so children which is concerned in the thesis is limited schooling ages from 6 to 18 years old child This definition will be used throughout the paper

Migration: General familiar definition of migration is the movement of human from one place to another According to International Organization for Migration, there is not the universally definition exists The term migrant can be applied for personal, family moving to another place for a better life without intervention of external forces Migration can be permanent, if that person will never return to his/her origin places, or for a long term he/she move to other places and live in that destination over at least 12 months A short term migrant is the person who moves to other place at least 3 months but less than

1 year Under definitions, traveler in short term or business persons will not be considered

as migrant However, migrant includes certain kinds of short-term migrant such as seasonal worker, farm-worker In the context of this paper, migrants are the people who work in another place over 6 last months and it doesn’t count permanent migrants who have permanent residence in other places different from origin

2.2 Theoretical literature

2.2.1 Lewis model (1954)

In Lewis’s theory, traditional agriculture and industrial sector exists simultaneously Because of the disparity between two sectors, if evaluate the productivity and human

6

resource of each sector, traditional agriculture which has surplus labor and low productivity will move in to industrial sector which has high productivity and high technology level Lewis shows that 30% of income difference between two sectors will make the labor move to the higher income For a long time, the gap of income between them is lower, the wage in agricultural sector will equal to the wage of industrial sector

At that time the attractiveness to promote the movement will be disappear

Although, theory of Lewis is considered as the basis model of development but it was also criticized by other economists Lewis thought that the relationship of reinvestment or capital accumulation and number of jobs is positive Clearly, when investing to the technology, the labor force will be reduced Moreover, Lewis shows the supply of labor in agriculture is elastic and limitless According to Todaro (2003), surplus of labor in rural is limited and quite small Thirdly, Lewis assumed that in the industrial sectors, wage will

be constant instead of increase over time In addition, diminishing return of industrial sector need to be argued, in while, in real the return of industrial sector also increase in some specific fields

2.2.2 The push pull theory of migration (1966)

Push pull theory (Lee, 1966) is neoclassical theory which explains why people move for rural-rural, rural- urban, urban-urban and international migration It emphasizes the trends of people move from low to high income area or from densely to sparsely population area or because of fluctuation of business cycle

Decision to migrate and process of migration can be summarized by 4 main factors: Characteristics of origin, characteristics of destination, intervening obstacles (cost, border, etc…) and nature of people Among them we separate it into two kind of factor, push factors include demographic grow low living standards, people lack of conditions to have

a better life such as economic opportunities, political repression, infrastructure, access to

7

clean water, sanitation, etc (in the origin area) Pull factors include demand of labor, land, economic opportunities, political freedom, good infrastructure in the destination and so on (in the destination) A potential migrant will consider the balance of all the attractive and interactive factors and attractive factors of destination along with difficulties of intervening obstacle and nature of human to decide migration

Some weak-points of theory could be given such as it couldn’t say why the reason to move It judged that economic growth is the main reason for people moving and unable to predict the movement in the future It also didn’t account the fact that some people have less ability to have migrating decision, people also differ in their ability to act of migrating decision although they want to migrate

2.2.3 Todaro model (Todaro, 1969)

Corresponding to macroeconomic theory is microeconomic, Todaro model is considered as a predecessor of theory for rural migration, Todaro model is the basic theory not only for internal migration but also international migration because of valuable cores Todaro explained that the persistent of internal migration is to face the unemployment in the destination places The existence of underemployment and unemployment of urban areas affects the probability of rural migration when they have difficulties in finding a job over there In contrast to other previous paper when considering labor transfer process include one- stage, Todaro showed that this process include two-stage First stage is the movement of unskilled rural worker to urban areas and spend time in one “urban traditional” sectors The second-stage is the phenomenon when there is the permanent movement in modern sector jobs

Although Todaro only concerned about internal migration but income hypothesis is also relevant with international migration The interdependence of mixing effect from industrial expansion, productivity growth, differential in earning capacity of urban versus

8

rural activities leads to the migration and occupational distribution in urban places When evaluating the determinant of urban labor supply it needs to concern about the rural- urban expected income Differential income will be adjusted so unemployment rate will

be a problem with urban supply when rural worker migrate to find a job with expected income From that it said that raising income by creation employment in rural area

s will help reducing out-migration in rural areas So the policy which government can support is development programs make rural life more attractive One more important matter affect the decision making of migrant is the balance between the risks and the probabilities of unemployed in one period with certain wage and the opportunities to have

a permanent job It also depends on the behavioral and spirit of migrant who prefer which fundamental role among them

2.2.4 Chain migration theory

Chain migration is the relationship between the previous migrants and current migrants through the movement of migrant from this place to other place follows previous migrants Previous migrants will establish the transportation, accommodation, information…so make it convenient for migrant to migrate Follows migrants path therefore, knowledge of destination increase, obstacles decrease so attractiveness of destination increase Lee’s (1966) argued that migration facilitates information of destination back to origin Moreover, Böcker (1994) emphasized migration network reduce the risks as well as material and psychological cost for next migrants

Chain migration and the expectation of migration may increase illegal immigration or migrant will find destination not as attractive as expected On the other hand, this theory doesn’t offer the insight into the mechanism which migration network can lead to weaken migration system It also didn’t indicate that counteract of external and internal processes lead to increase migration

of treatment group and comparison group The below equation represents a direct effect

of T on the program Y in context subject i with or without intervention:

Y_i= * X_i+ * T_i +ε_i Where T is the dummy variable with the value 1 if subject with intervention, 0 if subject without intervention

X: the observed characteristics of subject and other characteristics affect subject i ε: Error term representing unobserved characteristics which affect Y

The accuracy of impact evaluation depends on the reasonable assumptions as well as exogenous factors Khandker et al(2010) also represented some various economics methods to address selection bias and calculating the effect of interventions

2.2.4.1 Randomized evaluation method

Randomized evaluation identifies a group that has the same observed characteristics and assigning randomly interference to evaluate the effectiveness of one project or program Randomized evaluation method is used to address an array of well-known bias

It can resolve the selection problem that plagues treatment effect estimates Researcher is allowed to design behavioral parameters that are difficult to estimate by other methods The results of randomized evaluation are transparent and typically and highly credible to policymaker But randomized evaluation is quite rare because the way to collect data costly including expensive and take much time Moreover, it have little or no external

10

validity, usually estimated impacts are local Randomized evaluation also prone to biases

so misleading inference, and feasible in practice

2.2.4.2 Propensity score matching methods (PSM)

Propensity score matching estimates an average treatment from observable data and help to match treatment with similar control units observationally So to compare the intervention effects between treatment and control group which bases on the observed characteristics The key advantage of PSM is to use linear combination of covariates for a single score, so it can balance treatment and control group on a large number of covariates without losing a large number of observations But PSM only counts observable covariates only, all factors affect treatment but they can’t be observable, can’t be mentioned in the method Another issue is that PSM requires large sample so maybe overlap between treatment and control group From that, biases may be created after matching because of missing unobservable variables and overlapping data PSM has another disadvantage when assuming no unobserved differences and this is often implausible

2.2.4.3 Difference- in-Differences or double difference methods (DID)

Difference-in-Difference examines the effect of intervention of program by taking out the different between control and treatment group before and after intervening DID is easy to calculate standard errors by control variables which may reduce the residual variance DID eliminates fixed difference not related to treatment and can study treatment with different treatment intensity But DID method assume that the unobserved heterogeneity does not change in time but clearly when the trend change, it will make biased for the result In case unknown characteristics make control and treatment group react differently DID will easily overestimate or underestimate the true effect

11

Furthermore, DID require data which have 2 pre-intervention periods of data The lack of data can be applied DID method

2.2.4.4 Instrumental variables ( IVs) methods

Instrumental variables method proves its effectiveness and flexibility when allowing the selection bias on unobserved characteristics vary over time It can correct the bias through found instruments which impact on participation status but no relationship with unobserved characteristics IVs can control endogenous in the model effectively, thus it is usually used to address the problem of omitted variable bias, unobservable variable, errors

in variables or measurement errors of endogenous variable by permitting them remain in the error term But IVs can be deal with some problems when instrument correlate with errors term, lead to inconsistence More popularly, authors usually have to face with weak instruments and many confounders which results in imprecise and biased result Due to the first-stage equation, that instrument shows the weak predicting ability In the second-state, predict the outcome exactly is limited So the most important problem of IVs is how

to choose the good and suitable instrument variables

2.2.4.5 Regression discontinuity (RD) design and pipeline method

Regression discontinuity which is extent from IVs accounts for selection or participation on observed and unobserved characteristics Then, compare participant and non-participant groups which base on the closed location around criteria It can yield an unbiased estimate of treatment at the discontinuity RD also takes advantage when assigning the benefit as criteria for design of social policy But RD only looks at sub-group of sample and assignment rule, in practice, often not implemented strictly When estimating at the discontinuity, fewer observations exist to ensure not bias in the results

Pipeline comparison uses variation in the timing of program’s implementation to establish eligible comparison groups who is suitable to join the program but not receiving

2.3 Empirical literature

There has been not any quantitative research about the impact of migration and gender matters on child’s welfare in Vietnam But in the world, for years, many theories and empirical studies which were relevant to this impact

One paper published to quantify the impact of remittances upon school enrollment was a study of Cox-Edwards and Ureta (2003) on the risk of school dropout The authors used cross-sectional data from Encuesta de Hogares de Propósitos Multiples 1997 and estimated how remittances influence Salvadorian households’ educational The authors found out, via income effect, remittance relieve the budget constraints, it makes good condition for the household afford and allows children spend more time in school The result is that remittances reduce significantly the dropout aged 6 to 24 To be consistent with dropout decrease, authors expect the right thing is that remittances contribute positively to school enrollment However, one of concerns becoming the limitation of this paper is that the authors didn’t address potential sample selectivity issues and endogeneity

of remittances

13

Indeed, with the same hypothesis, Hanson and Woodruff (2003) used Mexico’s 2000 Census data and recognized that there was the complex interaction between remittances and migration Remittances and school enrollment have the positive relationship In Mexico, in migrant family, children (from 10 to 15 year-olds) complete more grades in school than others But the authors also noted that negative labor shocks which affected to parents would make the children have to spend more time to work instead of spending their time in the school It implied that there was the spurious negative relationship between migration and children schooling enrollment One concern is to overcome the limitation some previous paper, the authors knew to control the endogeneity of remittances by using IV with historical migration rates and household characteristics instrument

Acosta (2006) realized that remittances can affect child outcome and labor supply in

El Savador, the key outcomes for the growth of developing country The paper proved that the estimating of remittance was taken into account for selection and endogeneity problem which different from estimates presented in previous studies To solve relating problem, Acosta (2006) used Propensity score matching and instrumental variable method

to run regression, instrument variable which was used in the paper is migration network and household migration history (number of international migrants who returns two or more one year ago) for remittance receipt The result showed that remittance had positive relationship with school enrollment but not for all range of age, only for age from 11-14 (boys) and 11-17 (girls) year old Moreover, negative side of remittance is to increase of child work for wage However, concerning adult labor supply, the robust result showed that female labor supply is lower in recipients of remittance but higher with male labor supply

Mansuri (2006) analyzed the relationship of economic migration undertaken by men and it showed that children in migrant’s household attain more time in school and remain

14

in school with more accumulating years comparing to non-migrant household Mansuri (2006) used two strategies to address the endogeneity, the first strategy is to use instrument variables, the second strategy is to confine attention to migrant household and

to use information on the year of initial migration to exploit the fact which affect to children school enrollment With instrument variables, the paper found two instruments, one is migration network with a level village and another is the number of male adult in household with household level Discrimination happened in Pakistan because girls in migrant household headed by female were more likely to drop out of school than boys Moreover, both boys and girls in migrant household headed by female tend to work more than migrant household headed by male Education in Pakistan depended much on income flows in the household and traditionally structure of household and social culture

On differential impact of gender on wage income, expenditure and production of migrant-sending household, Pfeiffer and Taylor (2008) used probit-method and control potential endogeneity by using historical migration gender as instrumental variable They saw that impacts of migrant left behind for sending household according to the genders Male migration was more likely to engage production activities than female migration That was explained like the role of female migration in household production activities, they only participate in a little of work in a subset of production activities On the other hand, female migration had the negative relationship with education, female migration decrease the investment in education However, male migration had negative impact This was interpreted that the lack of monitor of female migration on schooling investment or the signal from international migrant about the low return of low skilled migrant

By contrast, migration can affect school enrollment negatively, one typical example, McKenzie and Rapoport (2007) recognized that migration and school enrollment of 16 to

18 year olds girls and 12 to 18 year olds boys in Mexico have significant negative relationship The argument of the authors is that migration affect school enrollment via

15

three ways: the income effect brought by remittances, direct effect of adult migration that requires more demand on child work, indirect effect of incentives to invest for education One specific thing of migration in Mexico is that children in migrant household likely to self migrate to US via illegal ways In this context, children in migrant household are unlikely better than children in non migrant household Once remittance became essential for survival of population in Mexico, it generated the dependency of household to remittances for household members who left behind Change in consumption, reduce labor supply, enjoy more time in relax and more need of remittance in the future The absent of adults in the household may increase the child’s responsibilities in the household, greater demand for older children to run and support the family It makes difficult to attend school like other normal children

To overcome some limitations in previous paper about El Savador, by using panel data and controlling household fixed effect method, Acosta (2011) also did another research and it found that female migration tend to reduce the child work in domestic and non-domestic But it was not true for male migrant whom seemed to stimulate the child work in domestic labor In addition, male migration seemed not to affect to school enrollment of child while female migration had the inverse impact when reducing the child school enrollment The reason for the negative impact of female migration on school enrollment maybe because of the absent of monitoring of female in managing funds or difference in the use of remittances by gender of the recipient person or the child-adult labor substitution

According to Nguyen T, and Purnamasari R (2011), the impact of international migration on sending household was affected a lot by migrant’s gender in context of Indonesia They used Family Life survey from 1993 to 2007 and instrumental variable method for regression the result As the whole, household with only female migrant reduced child work Surprisingly, household with some male migrants household seemed

16

to be insignificant with child work Interestingly, either migrant’s gender seemed not to

be significant with school enrollment of child, the result can be explained that the absent

of parent impact negatively in return to education, specially, women absent involving child care and monitoring children

Emigration can decrease or increase the school enrollment of children, the final result

of impact will depend on the level to offset the effect of disruptions happened in migrant household The effect of migration on children school enrollment equals to total sum effect of migration on children school enrollment via its impact on family income and family structure Example, migration can increase the school enrollment of children if remittances can be used for investing in education and that household remain a female household head to take care their children But if migration increases the migration decision of children in migrant household, it will lead to negative result with children school enrollment

This part of research builds up two important aspects relating to the migration and migrant’s gender Firstly, migration has the impacts on child work and children school enrollment of sending household To get the good results, it needs to be solved the endogeneity problem by using instrument variable Secondly, it determines that the migrant’s gender combines with migration can create above significant impacts

2.4 Conceptual framework

According to previous paper – Nguyen V.P (2011) which researched the way which female and male uses the international remittance in Vietnam, Nguyen V.P realized that shifting income from remittance affect intra-household decision making, gender of remaining adults in the household are likely to spend in different way Role of female increases when migrating to earn money, they have the power and contribute their voice

in the household But the absence of female adults create disconnects and loss of control

17

with children’s monitor So, impact of migrant’s gender on children’s school enrollment and child work base much on total sum of all aspects Through various results which are concerned in the literature review, it is expected the impact of migration and gender of migrant on children’s school enrollment and child work like below:

Figure 2.1: Conceptual framework about impact of migration and migrant’s gender on children’s welfare

The layout of conceptual framework advices that children’s school enrollment and child work can be impacted by migration and interaction of migration and migrant’s gender Net impact of migrant’s gender becomes ambiguous matter, depend on above concerning factors With migration factor, migration doesn’t always show its effectiveness when relaxing credit constraints, raising the living standard of all members

in the household Nguyen T and Purnamasari (2011) didn’t see relationship between migration and children’s welfare The net impact of migration also bases on the level of effect of each factor including remittance, the offset of presence of other family member and the absence of migrant in the household

if two variables are codetermined and affect to each other, endogeneity arises also Another reason like auto-regression which presents the output variable depends linearly

on its own previous value or auto-correlated which implies the cross-correlation of a signal with itself Consequence of this problem is to make regression coefficient of OLS biased Some popular methods are used to solve endogeneity problem like IVs and Heckman selection correction

3.2 Endogeneity of migration

After reviewing some mentioned literatures, it is realized that the problem we have to face in estimating effect of migration on children school enrollment and child work is the endogeneity There are likely to have the correlation between explanatory variable and error term by reasons of: reversed causality, omitted variables, self-selection leading to bias The relationship between children’s welfare and migration which may be happened

in two sides create reversed causality Children’s welfare can influence migration and

19

migration can influence children’s welfare reversely In schooling outcome case, in the household where it needs income to invest for education of children, so it demands family members to make decision for migration with the purpose to fund children’s education In child work case, to avoid depending on income from child work, the household must to migrate for living and get the better opportunity for financing their household’s economic Some variables which are so difficult to observe such as ability, connectedness or concern

of children’s education and child work result bias, they may lead to some overestimated effects such as positive effect for schooling and negative effect for child work outcomes The last issue of endogeneity is self-selection The migrant household can be selected randomly and sorted into migration basing on unobservable characteristics

The first rule when using instrumental variables method is to choose the best and fitted instrument variable Basing on the papers which concerned about endogeneity of migration and remittance in literature review, most of authors were likely to choose historical migration networks, household characteristics, distance… as instrument variables On the aspects of overview about the context of Vietnam, there is no data about

of distance from origin place to destination place of migrant Historical migration network

in destination to support the remaining people have more information, relax the budget, create more opportunities for migration easily And one more factor can be considered because it is so common with the reality condition: the number of male adult in the household Due to the immobilization and ability of female, they usually stay at home to take care of children If the household has more male adults, the higher of ability to migrate will be Clearly, two variables will affect directly to indicator of household migrant but not correlate to children school enrollment or child work So, in the first stage

of IVs method, household migrant indicator is considered as dependent variable It includes two instrument variables: historical migration network, number of male adults and other independent variables in second stage

20

3.3 Estimated equation

3.3.1 Validity of Instrument variable:

In place of migrant household variable, Nguyen, T and Purnamasari, R (2011) as well

as many Acosta (2006), Pfeiffer and Taylor (2008) used migration network history as a category of instrumental variable for endogeneity of migration The assumption in this instrumental method is that migration networks don’t affect children’s welfare but through the ability to have migrants in household Meanwhile, Mansuri (2006) used number of male adults such as instrument of migration The most important of impacts of migration network history and number of male adults are effects of human and physical capital intensity to induce more migration in the neighbor communities Mostly, migration network creates the social network in destination places and lower the cost of migrating via the support and information Number of male adults makes the household have more convenient conditions to distribute the labor in the household, therefore motivate migration In order to check the validity of instrument variable, we test endogeneity by Durbin Wu-Hausman test and then test the strength of instrument variable

by F-statistic

3.3.2 IVs methods

The instrumental variables method is used widely in econometrics and is one among the method to obtain the consistent parameter estimates

a) Inconsistent in Ordinary Least Square (OLS)

Consider one scalar regression model with dependent variable y and independent variable x When obtaining a consistent in estimating of a will gives the endogenous change in x Dependent and regressor variables are measured via deviations from regression model specifies:

21

y=a*x+u (2)

where in u is error term Regression of y on x yields OLS estimate a^ of a

Standard regression results make the assumption that error term not correlated with regressors There is only effect of x on y and error term on y like the path below:

x y

u

no relationship between x and u

But in some situations, there is the correlation between x and u, the standard to ensure the consistent in OLS is violated This phenomenon appears if there are some omitted factors or some measurement errors, auto-regression… in the model Error term affect x and make the impact of x on y higher level An appropriate diagram will be like this:

x y

u

x become have two effects on y The goal to evaluate the effect of x on y via a, now combine the effect of u also, giving a^ >a Using calculus, we have y= a*x+ u(x) with derivative:

dy/dx=a+ du/dx

From the giving information on dy/dx, the total effect which estimates on y is a+ du/dx rather than a alone The OLS estimations become inconsistent and biased for a

b) Instrumental variable

be ensured in the model to conclude that z is the instrumental variable The first assumption is z not to be a regressor of y, y is regressed by x The errors from missing z will be absorbed in the error term u At that time, z correlated with the error term The second assumption is there is the association between instrument and variable which is instrumented

The causal estimated calculation is estimated the change of dy/dz and dx/dz as:

aIV= 𝑑𝑦/𝑑𝑧

𝑑𝑥/𝑑𝑧

The obvious way to estimate dy/dz is to use OLS of y on z with slope estimate (z’z)

-1z’y Similarly, to estimate dx/dz we use OLS of x on z with slope estimate (z’z)-1z’x Detail like below:

aIV= (z’z)−1z’y

(z’z)−1z’x. =(z’x)-1z’y

23

This is called 2SLS estimators To find the 2SLS estimators, it is a good way to implement two-step procedures The idea of instrumental variable is to find a proxy for x which not correlated with error term u The first state will generate the proxy, the second stage will substitute the value of proxy for x and the result will be regressed by OLS The proxy must meet the requirement that it belongs in the second stage and predict the value

of x but doesn’t belong in first stage Other words, we have to find instrument z that impact directly on x and not impact on y Technical conditions on z as follows:

Testing for endogeneity can obtain by get the residual ε^ from the first stage then put ε^ in estimated equation (2), y=ax+δε^+u If δ^ is significant, so x has endogeneity To test weak instrument, obtaining F-statistics on the estimators of the instrumental variables

we run the first stage, check the test: Ho: δ1=…=δm=0 If F-statistic is high, instruments

24

are weak and 2SLS estimators will not be consistent or have large standard errors Thus,

in the problem where OLS break down because of the correlation of right-hand-side

variables and disturbances, IV is the good choice to get the consistent estimates

3.3.3 Estimated equation:

When estimating migration problem, it needs to be noted is endogeneity To solve

endogeneity, we can use instrumental variable (is also called two-stage least squares

2SLS) approach Historical migration network variable and number male adults in

household will be showed as instrumental variables to evaluate their impact on

migrationdirectly, and impact on the children’s welfare indirectly

From the literature review, basing on the theories and the model which authors build

to evaluate the impact of migration on children’s welfare, model which appeared in the

paper of Nguyen T, and Purnamasari R (2011) and Mansuri (2006), at the same time,

from the purpose of this paper to check about migration and characteristic of migrant can

affect to children’s welfare The estimating equation is for each household i:

M_i=β 1 + β 2 *X_i + β 3 *Migration_network + β 4 *No_male_adult + ε_i

(1- First stage)

Y_i=α 1 + α 2 *X_i+ α 3 *M^_i+ α 4 *female_i*M^_i + α 5 *t_migrant*M^_i + _i (2- Second stage)

In the first stage:

M: indicator for having a migrant

Migration network in 2008 will be applied as instrument variable for M in 2010 which

evaluated by the percentages of household in the village with migrants Historical

migration network has the meaning to create a useful network In destination, old

25

migrants have the experience, information, opportunity to have new job From their relationship between them and the remaining people, they will transfer the tool, knowledge and opportunity for migration in easier condition The lager migration network is, the more people migrate

Number of male adult is another instrument variable for M in 2010 which is evaluated

by the total number of male adult who over 18 year olds Male adults will affect directly to the ability of migration in the household The less male adult is, the less opportunity to migrate Meanwhile, male migrate to get more income for taking care the household

X_i: set of head household, household, village characteristics include log of per capita expenditure and other variables mentioned in section IV

ε_i : Error term

In the second stage:

Y_i: include children’s school enrollment and child work (children from 6 to 18 years old who also go to school or work in the last 12 months according to the 2010 VHLSS

X_i: set of household, head household, village characteristics mentioned in section section IV

t_migrant: Time of migrant living in the household in 1 year

Female_it: Dummy variable which show that households with only female migrant, omitted category can be referred as households with at least some male migrants

The paper put the interactive variables of gender of migrant and indicator of migrant, interactive variables of time of migrant living in household and indicator of migrant in

26

the model The reason can be explained that gender of migrant and time of migrant living in household will affect children school enrollment and child work via the occurrence of migrant

: un-observed components for household create the error

3.3.4 Method to run IVs regression

As the general estimated equation which concerned previous part, the detail equation can be written like below:

First stage:

H_migrant=β 1 + β 2 *HH_school + β 3 *HH_age + β 4 *HH_gender + β 5 *HH_work+

β 6 *ln_exp_pc + β 7 *Pro_females + β 8 *H_csize + β 9 *H_child + β 10 *Com_agri +

β 11 *Enterprise + β 12 *Big_road + β 13 *Electro + β 14 *Market + β 15 *Ele_school +

β 16 *Jun_school + β 17 *High_school +β 18 * Migra_net +β 19 *No_male_ad +ε_i

After running regression of first stage, H_migranthatregressor is predicted from H_migrant Then, we run second stage by putting H_migranthat and other relating variables in second stage model

Second stage:

Detail equation of children’s school enrollment can be written:

Enroll =β 1 +β 2 *H_migranthat + β 3* H_migranthat*female_it +

β 4 *H_migranthat*t_migrant+ β 5 *HH_school + β 6 *HH_age + β 7 *HH_gender +

β 8 *HH_work+ β 9 *ln_exp_pc + β 10 *Pro_females + β 11 *H_csize + β 12 *H_child +

β 13 *Com_agri + β 14 *Enterprise + β 15 *Big_road + β 16 *Electro + β 17 *Market +

β 18 *Ele_school + β 19 *Jun_school + β 20 *High_school + _i

27

Detail equation of child work can be written:

C_labor =β 1 + β 2 *H_migranthat + β 3* H_migranthat*female_it +

β 4 *H_migranthat*t_migrant+ β 5 *HH_school + β 6 *HH_age + β 7 *HH_gender +

β 8 *HH_work+ β 9 *ln_exp_pc + β 10 *Pro_females + β 11 *H_csize + β 12 *H_child +

β 13 *Com_agri + β 14 *Enterprise + β 15 *Big_road + β 16 *Electro + β 17 *Market +

β 18 *Ele_school + β 19 *Jun_school + β 20 *High_school + _i

Estimation strategy includes there steps:

Step 1: Running the first stage model and then replacing the predicted value of endogenous variable which calculated in first stage into second stage In detail, run first stage regression with dependent variable: indicator of migrant and independent variable: household head attributes, household attributes, village attributes and two main key instruments (historical migration network and number of male adults)

After checking the validity of endogeneity for migrant indicator and strength of key instruments in first stage, using predicted migrant indicator to create two variables which interacted between migrant indicator and migrant’s gender or time of migrant living in home After then, run second stage with dependent variables including children’s school enrollment and child work

Step 2: The paper will use diagnostic tests to check the appropriate model and avoid the biased coefficient and standard error Missing relevant variable or including irrelevant variable in model can lead to errors in the model and wrong evaluation Gujarati (2003) concerned about the strategy to build model by two ways: General to simple can make the coefficients less precise because of less variables However, the model still gets unbiased for all coefficients and hypothesis testing is still available Against, simple to general method can exclude the relevant variables out so it makes the biased, error in the model and hypothesis testing become invalid From that, most of the studies use general to

28

simple method In this paper, this method is superior to do diagnostic tests Begin with the least insignificant variables, we drop out one by one and re-estimate the model To be sure that the significance of remaining variables is improved, Wald test will be used to check whether we should keep or drop out the insignificant variables in the model

With cross-sectional data, one problem needs to be noticed is heteroskedasticity when existing non-constant error variance Heteroskedasticity violates OLS assumptions because least square estimator becomes insignificant and the results can’t be reliable We use Breusch-Pagan to test heteroskedasticity and use heteroskedasticity-robust to adjust the heteroskedasticity in the model

Step 3: After running regression and use diagnostic tests, make the report of final results and interpretation the results

3.4 Data

3.4.1 Source of data

Mainly data used in this research is Vietnam household living standard surveys (VHLSS) 2010 was conducted by General Statistics Office (GSO) with the technical support from World Bank The survey included the information relating to living standard

to make plan and policy for commune’s development in Viet Nam The sample was selected from 69.360 households in 3.133 communes/wards, including 22.365 households for income, 37.596 households for income and other issues, 9.399 households for income, expenditure and other issues, representative for rural and urban and regional levels Information was collected in 2 years: 2010 and 2011 through face to face interview The content of survey consisted of List of household member, Health, Income, Education, Expenditure, Durable goods, Accommodation

VHLSS 2010 is the last survey about wholly household living standard, contains information of current migration as well as household characteristics Information of

29

income in VHLSS is quite weak because there is no information about the value of housing and land ownership The advantage of this survey is to have the clear information for food and non-food items such as health, education, durables So in this paper, to replace the indicator for welfare we use per capital expenditure In addition to, VHLSS

2010 is rich in education and labor variables to help us determine the impacts of migration and migrant gender on children education indicators and child work indicators:

% of children 6-18 in the household who has gone to school, % of children 6-18 in the household who has worked in the last 12 months respectively

With detailed information of migrant gender in household, we can classify them into

2 groups: household with at least one male migrant, household with only female migrants

3.4.2 Variables description and measurement:

Standard to choose right migrant bases on the last movement which migrant live in rural areas and go far away from their households for working Time to go far away their households is over 6 last months Non-migrants are defined as people lives in rural areas, still live over there, and not move to other places for working in 6 last months

The dependent variables are the percentage of children of one household who has gone to school and has worked as labor force in last 12 months The independent variables include the social-economic factors which affect directly on the children’s school attainment and child work Among those variables, gender of migrant and indicator for having migrant are the most important variables

As above concerning about the non-economic factors which also affect to migration beside economic factors, it is classified into 4 sets of variables depends on the purpose using instrumental variable model The first set is head household characteristics, the second one is migrant’s characteristics alternatively, village characteristic are considered

30

also The last one includes the instrumental variables for indicator of migration historical network migration and number of male adults in the household

It is expected that in household characteristics will have positive impacts on children

if economy situation (Ln_pc_exp) can ensure the living standard, number of children (h_child, h_csize) in that household is small Perception of head household depends much

on age, gender, education (hh_age, hh_gender, hh_school, hh_work), such as young age, female head, high education level, much time working of household will ensure the children care better Migrant’s characteristics are the key impacts which we want to examine how the effect of these variables in this paper, including time of migrants living

in household (ti_migrant), gender of migrants (female_mi) and indicator for household migrant (h_migrant) The thesis also considers more about village characteristic, it is hoped that the presence of big road, electronic, market and health center ( Big_road, Electro, Market, Health_center) as well as the education system elementary, junior or high school (ele_school, jun_school, high_school) will make children’s welfare better The proportion people work in agriculture field (com_agri) brings the negative and the number of enterprise (Enterprise) in the villages has the positive relationship with children’s welfare hopefully At the same time, the thesis checks whether migration indicator is affected by the historical network migration (migra_net) and number of male adults (no_male_ad) Historical network migration and number of male adults will have positive relationship with migrant indicator as suggested

One among the most important variables is per capita expenditure which measures the welfare of the household for a long-term The variable reflects the welfares is not income because the shortage of income data in the household from the asset management system

as well as from the culture of Vietnamese, they don’t talk much with strange person about the real income However, as below histogram of distribution of per capita expenditure in

a year, it is not normal distribution and skews to the right