Determinants of secondary school dropouts in vietnam panel data evidence

ABSTRACT This paper investigates the socioeconomic determinants of dropout behavior of Vietnamese children in secondary schools using the Vietnam Household Living Standard Survey for 200

UNIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES

VIETNAM THE NETHERLANDS

VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

DETERMINANTS OF SECONDARY SCHOOL

UNIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES

VIETNAM THE NETHERLANDS

VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

DETERMINANTS OF SECONDARY SCHOOL

DROPOUT IN VIETNAM:

A PANEL DATA EVIDENCE

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

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

By

LE ANH KHANG

Academic Supervisor:

Dr LE VAN CHON

ACKNOWLEDGEMENT

Joining classes of quantitative research project with STATA & VHLSS2008, hold by the Faculty of Development Economics of HCMC University of Economics and the applied econometrics seminar by Prof Dr Ardeshir Sepehri from University

of Manitoba, Canada, have encouraged and yielded me confident to move this paper ahead

I would like to express my thanks to Mr Phung Thanh Binh, Mr Truong Thanh

Vu, Mr Nguyen Khanh Duy, Ms Ngo Hoang Thao Trang, and Mr Dang Dinh Thang and all other people participated for arranging and conducting the quantitative research project with STATA & VHLSS2008

I would like to express my appreciation to Dr Nguyen Hoang Bao who introduced the logistic regression model in explaining school dropout behavior at a

“sharing experience in doing research” seminar on August, 2010, hold by the Faculty

of Development Economics of HCMC University of Economics

I would like to express my gratitude to Prof Dr Ardeshir Sepehri who has sparked the idea of analyzing VHLSS dataset by panel data methods to capture the unobserved heterogeneity

I would like to express my sincere thanks to Dr Le Van Chon, my supervisor, who provides me directive suggestions during the thesis performing

I would like to thank all professors in the teaching board of MDE program, who have helped me accumulate valuable knowledge to acquire this study

To all my friends in MDE class 16, who give me emotional encouragements, I would like to express my thanks

Finally, I would like to express my deeply appreciation to my parents, to my wife and my son, to my family for spiritual supports In particular, I dedicate this thesis to my father

ABSTRACT

This paper investigates the socioeconomic determinants of dropout behavior of Vietnamese children in secondary schools using the Vietnam Household Living Standard Survey for 2006 and 2008 and logistic regression model for Panel data Determinants are considered at individual, household, schooling, and regional levels

My findings reveal that the unobserved individual characteristics account for 17% in propensity of dropping out of secondary schools in different years 2006 and 2008 Furthermore, the results disclose that child gender, child age, child ethnic, child inactive days, household expenditure, household head gender, household head education, the number of children between 1 and 17 years old, cost of school, urban-rural, and regions have statistically significant relationship with secondary school dropout

Key Words: secondary school dropout; panel; logistic model; random effects;

Vietnam

TABLE OF CONTENTS

CHAPTER 1:INTRODUCTION 6

CHAPTER 2:LITERATURE REVIEW 8

2.1 A standard model of household schooling investment decision 8

2.2 Empirical studies of school dropout in the world 10

2.3 Empirical Studies of school dropout in Vietnam 12

CHAPTER 3:VIETNAMESE SECONDARY EDUCATION – AN OVERVIEW15 CHAPTER 4:RESEARCH METHODOLOGY 20

4.1 Data 20

4.2 Methodology 22

CHAPTER 5:THE RESULTS 29

5.1 Descriptive Statistics 29

5.2 Dropout rates and Children characteristics 31

5.2.1 Dropout rates and Household characteristics 32

5.2.2 Dropout rate and School characteristics 35

5.2.3 Dropout rates and Regional characteristics 36

5.3 Regression Results 37

CHAPTER 6:CONCLUSION 45

REFERENCES 47

APPENDIX 52

LIST OF TABLES

Table 4.1: Description of the variables 26

Table 5.1: Descriptive statistics 30

Table 5.2: Regression results of the Random-effects models 37

Table 5.3: The estimation of dropout probability, given initial probalibity P0 38

LIST OF FIGURES

Figure 3.1: Secondary school dropout rates 15

Figure 3.2: Gross enrollment rate by urban-rural 16

Figure 3.3: Gross enrollment rate by gender 17

Figure 3.4: Gross enrollment rate by region 17

Figure 3.5: Average expense on secondary education per schooling person in the past 12 months by urban-rural 18

Figure 3.6: Average expense on secondary education per schooling person in the past 12 months by gender 18

Figure 3.7: Average expense on secondary education per schooling person in the past 12 months by gender 19

Figure 3.8: Average expense on secondary education per schooling person in the past 12 months by income quintile 19

Figure 5.1: Dropout rate and child age 32

Figure 5.2: Dropout rate and household expenditure quintile 32

Figure 5.3: Dropout rate and years of schooling of household head 33

Figure 5.4: Dropout rate and number of children between 1 and 17 years old 34

Figure 5.5: Dropout rate and cost of school 35

Figure 5.6: Dropout rate and region 36

LIST OF ABBREVIATIONS

GSO: General Statistics Office

HH: Household

LMP: Linear Probability Model

MOET: Ministry of Education and Training

MP: Maximum Likelihood

NA: Not Applicable

OR: Odds Ratio

RE: Random-effects

VHLSS: Vietnam Household Living Standard Survey

CHAPTER 1: INTRODUCTION

During the past two decades, Vietnam has achieved important results in education

in terms of increased enrollment, improved school infrastructure and diversified schooling forms (MOET, 2006) However, Vietnam is still facing critics on the quality

of education and struggling with the phenomenon of dropping out of school The net enrollment rates1 were 95.5% at primary level, 82.6% at lower secondary, and 56.7% at upper secondary level (GSO, 2011) The effects of school dropout are expounded in the costs of individual, community, and society Specifically, individual faces risk in finding jobs; country struggles low-skilled labor force; and society expands rich and poor gap These effects have raised concerns to many researchers around the world in examining factors affect school dropout, and from that appropriate policies are proposed to policy makers to find ways to mitigate the phenomenon

There are many factors which could influence early dropping out of school Empirical studies point out four groups of influential factors: individual characteristics, household characteristics, school characteristics, and regional characteristics However, most of the empirical studies in Vietnam utilized the cross-sectional data to examine the effects of these factors on the dropout behavior (Behrman & Knowles, 1999; Vo Thanh Son et al., 2001; Vo Tri Thanh & Trinh Quang Long, 2005; Nguyen Linh Phuong, 2006) Cross-sectional data face the possible problem of heteroskedasticity, specifically the unobserved individual effects Panel data are advocated to control for this

By addressing above issues, in this research, I am aiming at using panel data, rather than cross-sectional data, to examine the influences of the socioeconomic

1 Net enrollment rate at z level is the number of pupils who in the ages of z level (according to the education law

in 2005) and currently keep schooling at z level as a percentage of z level aged population Where, z is primary or lower secondary or upper secondary For example, if z is primary level then the net enrolment rate at primary level is a percentage of the number of pupils who aged from 6 to 11 years old and currently keep schooling at primary level over the number of primary level aged population

determinants on the dropout behavior of pupils in secondary schools in Vietnam with the aid of logistic regression model

My study is endeavored to achieve three main objectives: (1) To determine factors theoretically affecting the decision of dropping out of school; (2) To examine factors statistically explaining the dropout behavior in secondary schools in Vietnam; and (3)

To implicate ways to reduce the secondary school dropout rates in Vietnam The main question of the research is: “What are the determinants of secondary school dropout in Vietnam?” To answer this question, I divide it into two subquestions: (1) What are the determinants theoretically influencing the decisions of dropping out of school? (2) Do these determinants statistically explain the dropout behavior in secondary schools in Vietnam? The first subquestion will be answered by recalling literature review and empirical studies in the world and in Vietnam Determinants obtained by the first subquestion will be used for the second one by applying econometric method to analyze the secondary data VHLSS

The paper is continued with a set of sections Section II recalls the literature review and empirical studies in the world and in Vietnam Section III provides an overview of education system in Vietnam with a brief picture of dropout situation during 2000-2006 Section IV describes the dataset used and research methodology Section V presents the results based on descriptive statistics and econometric models Section VI comes up with main conclusions and policy implications

CHAPTER 2: LITERATURE REVIEW

The issue of school dropout has attracted numerous researchers around the world The starting point to understand the decision of dropping out of school is the standard theory on human capital investment, originally developed by Ben-Porath (1967) and Becker (1964) The theory states that benefits and costs generated by additional schooling, e.g future income improving; expenditure on schooling tuition; opportunity costs of entering the labor market late, etc., will be compared by individuals If the marginal rate of return to additional schooling exceeds the marginal cost of education, individuals will keep schooling The limitation of this theory is the assumption that individuals face no resource constraints This assumption does not seem to hold in reality Moreover, dropping out of school is not individual decision Children don’t decide by themselves but mostly by their parents The household schooling decision theory releases the assumption of no resource constraints and considers an existing relationship between parents and children, in which parents play a principal role and children as an agent In parents’ view, children’s education is considered as both consumption goods and investment goods Parents spend resource on children’s education because well-educated children bring satisfaction to them Parents invest in children’s education with the hope that they will receive support from children later in life A standard model of household decision-making in terms of children’s education

is discussed in detail in a paper by Vo Tri Thanh and Trinh Quang Long (2005) In this section, I would like to briefly recall this model and also underline empirical papers in the world and in Vietnam related to the main implications of the model

2.1 A standard model of household schooling investment decision

The household schooling investment decision model begins with an assumption

that households are considered as unitary households It means there is no difference in

preferences of parents If parents’ preferences are not the same, they are then supposed

to behave as if they are maximizing a single utility function

Suppose a household includes a father, a mother, and N children, in which N children are divided into m daughters and n sons The parents’ life is divided into two periods They work and raise children in the first period They retire in the second period In the first period, income from working after subtracting a proportion of investment in their children’s education is considered as household consumption In the second period, their consumption depends on the remittances that their children return

to them The amounts of remittances in turn depend on the level of children’s education acquired in the first period Hence, there is a trade-off in parents’ schooling decision between consumption in the first period and consumption together with children’s wealth in the second period A utility function which represents the identical preferences of parents is expressed as follows:

) , , , , , , , (C1 C2 I d1 I dm I s1 I sn

U

Where, C1 and C2 are household consumption in the first and second periods respectively I di (i = 1 … m) and I sj (j = 1 … n) are incomes earned by the ith

daughter and jth son in second period respectively

The equation 2.1 can be expressed in another form as follows:

) , , , , , , ( ) (C1 G C2 I d1 I dm I s1 I sn

F

By a series of arguments which we can see in Vo Tri Thanh and Trinh Quang Long (2005), the demand for quantity of daughters’ and sons’ schooling are pointed out as a function of the cost of education, parents’ wage rates on the labor market, children characteristics, unearned income, parents’ education, household and community factors as follows:

),,,,,,,,

S

),,,,,,,,

S

Where,

S (i = 1… m) and S sj (j = 1… n) are the education of the ith daughter and jth son

m

w and w f are mother’s and father’s wage rates respectively

V is unearned income such as parent’s satisfaction from well-educated children

P is the direct cost of education, including tuition fees, books, uniforms, etc

m

S and S f are mother’s and father’s education respectively

di

Z and Z sj are daughter and boy characteristics respectively

H is other household and community factors

The implications of the model are the wage rates of parents on labor market, the unearned income, the cost of schooling, the parents’ education, the children individual characteristics, and other household and community factors do have effects on the parental decision of investing in their children’s education (for more details, see Vo Tri Thanh & Trinh Quang Long, 2005)

2.2 Empirical studies of school dropout in the world

The determinants of dropping out of school are empirically divided into four groups of factors: (1) individual characteristics; (2) household characteristics; (3) school characteristics; and (4) regional characteristics

Numerous papers examine the influence of individual characteristics on the dropout decision The returns to education of boys, predicted by parents, are higher than those of girls (Schultz, 1993) Therefore, benefits of investing in education for girls may be lower than boys Moreover, in developing countries and in rural areas, the opportunity cost of educating girls is higher than that of boys as girls are supposed to perform more household works than boys As consequence, the demand for girls’ schooling will be lower (Glick & Sahn, 1998) The older the age is the greater the tendency of school dropout Children in working age tend to engage in the labor market

to assist their parents Older children often go along with higher opportunity costs and lower marginal benefits could discourage parents from investing in education for them

(Ben-Porath, 1967) Child labor has a positive relationship with school dropout (Admassie, 2002) Working absorbs much of children’s time instead of using it for schooling Energy exhausted from labor affects children’s performance at school Children with poor mental health have a positive relationship with dropout status There are a few researches on how health issues directly affect school dropout (Pridmore, 2007) But in general, researches indicate that poverty often results in poor health and under-nutrition Through there, children’s educational access and attainment are severely jeopardized (Glewwe & Jacoby, 1995; Alderman et al., 2001; Grira, 2001; Ghuman et al., 2006)

Household characteristics are empirically considered as important determinants of dropping out of school Children, whose parents have difficulties in finance, are more likely to drop out of school Children in low income families are heavier affected in terms of school completion than children in high income families (Duncan et al., 1998; Glick & Sahn, 1998) Children with more educated parents are less likely to drop out of school than ones with less educated parents (Glick & Sahn, 1998) Children receive more supports in learning from their educated parents would help them stay in school longer (Sabates et al., 2010) Specifically, mother’s education has a stronger effect on children in school than father’s education Moreover, the effects of parents’ education

on school dropout are different between boys and girls The dropout probability on girls is larger than boys (Tansel, 1997) The higher the number of children is, the larger the probability of school dropout, because of the budget constraints on families (Psacharopoulos & Arriagada, 1989) In households which have higher number of children, the dropout probability on girls is higher than boys (Parish & Willis, 1993) due to raising demand on girls’ childcare

School characteristics are factors that may influence school dropout Many studies show that high schooling fees increase the probability of school dropout (Wolfe & Behrman, 1984; Chernichovsky, 1985; Al-Samarrai & Peasgood, 1998; Zimmerman,

2001; Dostie & Jayaraman, 2006) Distance to school plays an important role in developing countries as it may force children to discontinue education, especially in rural areas where most of schools are located far from children’s houses, and the means

of transportation are not well developed (Bilquees & Saqib, 2004) Poor education quality discourages children to remain in school and parents’ motivation to keep their children schooling (Coleman, 1966; Oakland, 1986a & 1986b; Brown & Park, 2002; Hanum, 2003; Hanushek et al., 2006)

Regional characteristics are another factors affect school dropout The dropout probability is higher for children living in rural areas than in urban areas (McCaul, 1989; Ono, 2000) School enrollment is significantly affected by geographic disparities Poor regions tend to have higher school dropout rates (Vo Tri Thanh & Trinh Quang Long, 2005)

2.3 Empirical Studies of school dropout in Vietnam

Empirical studies in Vietnam find that all above four groups of factors have strong influences on the school dropout

Behrman and Knowles (1999) estimate the relationships between household income and the school success of children in Vietnam by utilizing data from the 1996 Vietnam Social Sector Financing Survey They find that there is five times higher in the income elasticity of completed grades compare to the median estimate of earlier studies For grades completed per year of school, this relationship is even strongest Moreover, between boys and girls, this association is quite difference This difference suggests that girls’ schooling is considered to be more luxury than boys’ schooling Furthermore, the paper indicates that school fees are only one-third of what households directly consume on education Thus, school fees exemptions are necessary but the influence in school enrollment by this exemption is not widely

Vo Thanh Son et al (2001) utilize the data from VHLSS 1998 to explore variables associated with dropping out of school Some evidences are provided from

their study: (1) children from households in the poorest quintile have higher dropout rates compared with ones from households in the top quintile; (2) an increase in school fees leads to increase in dropout rates; (3) gender matters are not evidentially at primary level, but it does appear in secondary level Specifically, it becomes wider at upper secondary level, compared to the lower one This is considered as evidence that girls tend to have higher dropout rate than boys

Vo Tri Thanh and Trinh Quang Long (2005) identify the underlying determinants

of the schooling dropout in Vietnam by separately using data from three Vietnam’s Living Standard Surveys conducted in 1992/93, 1997/98 and 2001/02 They explore that: (1) the household’s per capita expenditure and the direct costs of school have strong effects on the dropout probability; (2) when household's per capita expenditure

on girls increases, girls would benefit more than boys But they would suffer more than boys from an increase in the direct costs of school However, these differences have been gradually narrowed substantially; (3) dropout phenomenon is a regional specification Different regions have different effects on dropout rates; (4) dropout situation is very much dependent on the public funding for education

Nguyen Linh Phuong (2006) investigates the effects of parental socioeconomic status, school quality, and community factors on the enrollment and achievement of children in rural areas in Vietnam by using VHLSS 1998 data The paper reveals that: (1) the levels of household expenditures and parental education have significant impacts on educational enrollment and outcomes Especially, mother's education is more important in determining school enrollment than educational outcome, while father’s education expands the probability of learning; (2) the dropout probability of girls is higher than boys; (3) school fees do not determine school enrollment as the exemption or reduction in these fees already applied to many of children in poor families

Le Thi Nhat Phuong (2008) utilizes VHLSS 2004 and 2006 separately to examine the socioeconomic determinants of school dropout for Vietnamese children aged 11-18 She finds that: (1) age and household size have significantly positive effects on the dropout probability; (2) the dropout rates are also shown to vary between girls and boys, but this gender gap has narrowed substantially Moreover, minority girls confront more obstacles in remaining in school than minority boys; (3) the school dropout rate is also very sensitive to the changes in household’s income and costs of school However, the costs of school have different impacts on families in different quintiles; (4) region

is another determinant affecting children’s decision to drop out of school; (5) the parents’ perception of the value of education may increase the child’s probability of school retention

Ngo Hoang Thao Trang (2010) employs VHLSS 2006 to examine the effects of individual, household, community, and regional levels on the dropout behavior of children in secondary schools in Vietnam by using the logistic regression model She finds that: (1) age, working hours per year, parents’ education, regions have large effects on the probability of leaving school; (2) household expenditure, the number of siblings, the proportion of pupils with reduced contributions, the pupil to teacher ratio, the pupil to classrooms ratio and the proportion of classrooms with good blackboards have small effects on the probability of leaving school However, the existence of children’s working hours per year in her model might leads to incorrect standard errors and inefficient estimation because of causal relationship between child work and school dropout

CHAPTER 3: VIETNAMESE SECONDARY EDUCATION – AN OVERVIEW

The Vietnamese national educational system is regulated by the Education Law (2005) According to this law, educational levels include general education with primary education, lower secondary education, and upper secondary education Primary education is conducted in

05 years of schooling, from the 1st to the 5th grade, where the age of commencement to the 1st class is six Lower secondary education requires 04 years of schooling, from the 6th to the 9th grade Pupils entering the 6th grade must complete the primary education programme, at the age of 11 Upper secondary education is conducted in 03 years of schooling, from the 10th

to the 12th grade Pupils entering the 10th grade must have a Lower Secondary Education Diploma, at the age of 15 In this section, I focus on lower secondary and upper secondary educations to graphically provide a brief picture about dropout trends during the past period 1999-2011

The MOET (2011) reported statistical data on education from 1999 to 2011, in which the dropout rates were calculated but only available from 1999-2000 to 2004-

2005 (Appendix 1) Figure 3.1 provides dropout rates in secondary schools The dropout rate decreased from over 8% in 1999-2000 to around 5% in 2004-2005 at lower secondary education, while it increased at upper secondary education from over 7% in 1999-2000 to above 8% in 2004-2005

Figure 3.1: Secondary school dropout rates

0 2 4 6 8 10

However, according to Vo Tri Thanh and Trinh Quang Long (2005, p.25), this data contains the issue of underestimating of school dropout in Vietnam because it does not count pupils who stop schooling after completed a given grade Given that issue, a more precise definition are introduced by considering a child to be dropped out if he/she did not enroll in school in the 12 months prior to the survey, although this definition still faces the issue of ignoring a small number of children who postpone their education in the 12 months prior to the survey but intended to return to school in the coming years

Figure 3.2: Gross enrollment rate by urban-rural

Upper secondary

Lower secondary

Upper secondary

Lower secondary

Upper secondary

Source: GSO (2006, 2008, 2010)

GSO of Vietnam issued the reports on education based on VHLSS2006, VHLSS2008 and VHLSS2010 (Appendix 2) Some figures from these reports are quoted here to briefly provide a picture of secondary education in Vietnam in recent years Figures 3.2 provides gross enrollment rate vary across urban-rural The gross enrollment rates were not varying much in lower secondary level compared to upper secondary level

The gross enrollment rates, distinguished by gender are revealed in figures 3.3 There was not much disparity in lower level, while female had higher enrollment rates than male in upper secondary level

Figure 3.3: Gross enrollment rate by gender

0 20

Upper secondary

Lower secondary

Upper secondary

Lower secondary

Upper secondary

Figure 3.4: Gross enrollment rate by region

Red River Delta North East North West North Central Coast

South Central Coast Central Highlands South East Mekong River Delta

Source: GSO (2006, 2008)

Average expenses on secondary education per schooling person in the past 12 months, divided by urban-rural are provided in figure 3.5 Expenses on secondary education were double in urban area compared to rural area

Figure 3.5: Average expense on secondary education per schooling person in the

past 12 months by urban-rural

0 500

Upper secondary

Lower secondary

Upper secondary

Lower secondary

Upper secondary

Figure 3.6: Average expense on secondary education per schooling person in the

past 12 months by gender

0 500

Upper secondary

Lower secondary

Upper secondary

Lower secondary

Upper secondary

Source: GSO (2004, 2006, 2008)

Average expenses, divided by region are disclosed in figure 3.7 Expenses were highest in South East region while North West region got the lowest

Figure 3.7: Average expense on secondary education per schooling person in the

past 12 months by gender

Figure 3.8: Average expense on secondary education per schooling person in the

past 12 months by income quintile

0 500

CHAPTER 4: RESEARCH METHODOLOGY

Data which are repeated measurements on the same individual at different points

in time are called panel data or longitudinal data In this research, I am using panel data

to examine the school dropout phenomenon at secondary level from grade 6 to 12 Only children age from 11 to 18 years old who have accomplished primary level and keep schooling in 2006 are considered Additionally, children at 18 years of age who have accomplished upper secondary level are eliminated out of the sample For example, in 2006, a child at 11 years old who finished primary level will be considered

in 2006 sample In 2008, he/she is at 13 years old will continuously be included in

2008 sample There are some issues in the sample that I need to point out Some children who 11 years old in 2006 but 14 years old in 2008, or who 17 years old in

2008 but accomplished upper secondary level, or who 17 years old in 2006 but 18 years old in 2008 Why is that? The answer is because age is calculated to age rounded

at the time of the survey For example, a child who born in 1994 but in the month after the surveyed month in 2006, so he/she is not enough months to be recorded as 12 years

old, but 11 years old instead At the survey time in 2008, the surveyed months are after the born month Hence he/she is enough months to be recorded as 14 years old Similar

to children who are 17 years old in 2008 but already finished upper secondary level These children are also eliminated out Only children satisfy above conditions in 2006 are maintained to see whether they drop out of school in 2008, when they are reinterviewed

The panel data help us capture factors effect children’s dropout behavior during period 2006-2008 better than cross-sectional data In detail, suppose that we consider a cross-sectional dataset in 2008 only Children who are considered as dropout are evaluated through the question asking whether he/she is schooling in previous 12 months Suppose there is a child not schooling in previous 12 months then he is considered as dropped out in the sample The problem is that we are not sure whether this child just left school in previous 12 months or long time ago, say 24 months or 36 months ago Then factors that we base on such as his household expenditure, cost of schooling, … which also be captured in the period of 12 months prior to the survey, are wrong in analyzing the effect By using panel data, we assure that all children are schooling in 2006 and dropout decisions only happen in period 2006-2008 Hence, factors which effect dropout decisions are better evaluated

After consolidated to remove errors and inconsistencies, the sample data remains 1,869 children from ages 11 to 17 in 2006 and 12 to 18 in 2008, who are both interviewed in 2006 and 2008, in which 1,610 children keep schooling, 259 children being dropped out The dropout rate was generally around 13.86%

4.2 Methodology

I am concerning on the probability of secondary school dropout over two-year

periods, specifically, the determinants of the probability p of the occurrence of the school dropout rather than non-dropout that occurs with probability of 1-p In regression analysis, I want to measure how the probability p varies across children as a

function of regressors Hence, dependent variable is a binary outcome There are two standard models for binary outcome: the probit model and the logit model These models are fitted by maximum likelihood (MP) A linear probability model (LPM) which is fitted by ordinary least squares can also be used However, LPM faces problems, e.g., heteroscedasticity, the difficulty of interpreting probabilities which greater than 1 and less than 0, constant marginal effects (Gujarati, 1995) Then, Logit and Probit models are alternative choices Many papers choose logit model because of its mathematical simplicity (Gujarati, 2003) In this study, Logit model is employed After decided logit model is analysis model, the question is which estimator is suitable to treat this nonlinear panel model? One way is to explore the variations of time-varying regressors If between variation is the most variation rather than within variation, then FE estimator is not expected to be very efficient as it builds on within variation only Other way is using appropriate statistic software to fit the model by FE estimator If the outcome is “convergence not achieved” or “substantially larger standard errors” because of the loss of time-invariant observations and only within variation of the regressors is used, then FE estimator is not appropriated and RE estimator is alternative selection The point is that I want to eliminate the endogeneity

of dropout decision by randomly assigning this to children Then each child will have

an autonomous dropout decision which is not correlated with dropout determinants and this autonomous dropout decision is random and follows a normal distribution with

mean = 0, variance = squared of sigma_u2 In other words, the RE logit model designates the intercept to be normally distributed RE estimator can be obtained by xtlogit command to fit the RE logit model for whether children drop out of secondary schools

The Logit model:

i ki k i

y* 12 2   

Where, y i * is unobservable variable, but y i = 0 if y i * < 0 and y i = 1 if y i * ≥ 0

)

(

)

( ) 0 ( ) 1 (

2 2 1

2 2 1

*

ki k i

ki k i

i i

i

x x

F

x x

u P y

P y

Where, F is the cumulative distribution function of u i We are assuming that the

probability density function of u i is symmetric

In the logit model, we assumed that u i has a logistic distribution The probability

density function of u i is given by 2

) 1 ( ) (

i i

u u i

e

e u

f



The model is estimated by Maximum Likelihood Estimation (MLE) It can be

verified the cumulative distribution function (CDF) of u i is

) 1 ( ) (

i i

u u i

e

e u

ki k i

x x

x x ki

k i

ki k i

i i

i

e

e x

x F

x x

u P y

P y

2 2 1

*

2 1

2 2 1

1)

(

)

()0()1

(

The probability of non-dropout of school P (y i =0):

ki k i ki

k i

ki k i

x x x

x

x x i

e e

e y

2 1

2 2 1

1

11

1)0

(

ki k

x i

)0

(

)1

(

2 Sigma_u is the standard deviation of the random effect It can be obtained in Stata output

Or i k ki

i

i

x x

y P

) 1 (

i i i

i

O P

P y

P

y P

)1(

is the Odds Ratio (probability of dropout over

probability of non-dropout),

Or the log of the odds ratio that y i = 1 is a linear function of the explanatory variables

Hence, 100 x β k then has the interpretation as the % increase in the odds ratio due

to a one unit increase in x k , or a one unit change in x k results in a β k unit change in the log of the odds ratio, given other things, or the probability of the dropout changes ek The marginal effects in the logit model:

2 2 1

2 2 1

ki k i

ki k i

x x

x x

ki ki

i

e

e x x

) 1

2 2 1

ki k i

ki k i

x x

x x k

i

k P P

The sign of k is the sign of marginal effect

Intensity of the marginal effect:

Suppose that with x ki we have P i = P 0 When x ki increases x k(i+1) then we can calculate P i = P 1 from initial P 0 as follows:

ki k

x

e P

2 2 1

1

) 1 (

1

1 1

2 2 1



k k

e O

e O

P 



0

0 1

1 



Empirical Model:

The Sample Regression Function (SRF):

i i i

i i

i

i

e X X

X X

^ 3 2

^ 2 1

^ 1

^ 0

Variable Codings: Table 4.1 defines variables which are applied in the paper

Table 4.1: Description of the variables

dropout NA

The secondary school dropout is a binary variable which takes two values: 1 if dropout; 0 if non-dropout This is an observable variable In VHLSS, there is a question asking about whether a child is schooling in previous 12 months prior to the survey If the answer is NO, then dependent variable takes the value of 1; if YES = 0

Independent variables: Independent variables are divided into 4 groups

(1) Individual characteristics

chgen (-) Child gender is a binary variable; chgen = 1 for boy; 0 for

girl Less dropout probability is expected for boys

chage (+)

Child age is a continuous variable, be measured by years old The older the child is, the higher the probability of school dropout

ethnic (-)

Ethnic is a binary variable; ethnic = 1 for Kinh Hoa; 0 otherwise Children belong to Kinh Hoa ethnic group are expected lower in dropout probability

Child work is a continuous variable, be measured by the total number of working hours per year of a child The more the working hours per year of the child is, the higher the probability of school dropout

chill (+)

Child ill is a dummy variable; chill = 1 if he/she is illness/injury in the 4 weeks prior to the survey 3 ; 0 otherwise This variable is used to measure the health status of a child Children with poor health status are expected with higher dropout probability

chinadays (+)

Child inactive days is a continuous variable, be measured

by the number of inactive days (have to stay at home, not schooling) in the 4 weeks prior to the survey This variable

is used to capture the severity of the illness The more the severity of the illness is, the higher the dropout

probability

3 The use of 4-week data rather than 12-months data is due to its shorter recall period (Sepehri, et al., 2006)

(2) Household characteristics

loghhexp (-)

Log of household expenditure is a continuous variable, be measured by taking log of total real expenditure per year (in thousand VND) of the household Expenditure is used instead of income because income is usually not accurately reported The higher the household expenditure is, the lower the probability of school dropout

hhgen (+)

Household head gender is a binary variable; hhgen = 1 if household head is male; 0 female Children with male household head gender are expected with higher dropout probability

hhmar (-)

Household head marital status is a binary variable; hhmar

= 1 if married, 0 otherwise Children with married household head are expected with lower dropout probability

hhedu (-)

Household head education is a continuous variable It is the finished grade of the household head if his/her highest accomplished level is upper secondary education Else household head education equals 15 years if college level;

17 years if bachelor level; 19 years if master level; 22 years if PhD level Children with higher household head education are expected with lower dropout probability

nochi (+)

Number of children in a household is a continuous variable, be measured by the number of children aged from 1 to 17 in household Children in household with higher number of children (1~17) are expected with higher dropout probability