Essays on public policy reform impacts on vietnamese household outcomes

In the last decade economic growth among developing nations has been especially rapid. However, as countries continue to grow public policies and public policy administration must keep pace with the needs of the society they intend to support. National progress requires intensive investigation and accurate identification of policy issues so that policy makers may effectively plan and implement changes. In terms of economic progress, Vietnam is an example of how economic development, poverty reduction and human capital accumulation go hand in hand. After determining that human capital and household development were vital to Vietnam’s continued success, policies in these areas began to be proposed and adopted. This has led to human capital accumulation, improved household economics, and benefits in health and education sectors. This dissertation seeks to quantify some of these benefits and examine to what degree the success is attributable to policy changes. Using a quasiexperimental differenceindifferences approach, with propensity score matching, this manuscript combines three essays that isolate public policy reforms in various sectors and measures the outcomes. The first essay examines the effects of national health insurance reform on children’s educational outcomes. Households in the state sector were unaffected before and after the reform, and so children in that group served as a natural control group, whereas children growing up in nonstate employed households formed a suitable treatment group. Educational outcomes were measured for the three levels of general education: primary, secondary and high school. Results showed that the national health insurance reform improved educational outcomes for children in high school, both in terms of enrollment and school completion likelihood. Furthermore, it was shown that children from minority groups, females, those in rural areas, and those from poorer families were less likely to derive the same educational outcomes when compared to their counterparts. These findings are the first of their kind using the Vietnam household living standard survey data and would be of value to policy makers in countries that plan to adopt a similar health policy. The second essay was extended findings from the first to evaluate the impact of the national health insurance reform in 2005 on the household consumption. It exploits the difference between households in the state sector (control group) and households in

逢 甲 大 學

經 濟 學 系 博士論文

公共政策變革對家庭決策之影響

Essays on Public Policy Reform: Impacts

on Vietnamese Household Outcomes

指導教授：郭祐誠

中 華 民 國 一 百 零 九 年 七 月

逢 甲 大 學

經 濟 學 系 博士論文

公共政策變革對家庭決策之影響

Essays on Public Policy Reform: Impacts

on Vietnamese Household Outcomes

指導教授：郭祐誠

中 華 民 國 一 百 零 九 年 七 月

con-I also offer my sincere thanks to the members of my review committee, Professors Hung-Lin Tao, Ho-Don Yan, Sheng-Jang Sheu, Chi-Yin Wu, and Jia-Huey Lin Their comments, thoughts and ideas have improved this paper in numerous ways I would also like to acknowledge the faculty members and staff in the Department of Economics

at Feng Chia University They were consistently kind, patient, and quick to extend ever ever assistance they could throughout the many phases of this research

what-Finally, I wish to mention the enduring patience of my family Words cannot quately express how grateful I am for the support of my mother Hoang Thi Keo, father Phuong Huu Ai, wife Mong Thi Nguyet and my two sons Phuong Minh Khoa and Phu-ong Minh Duc My two older sisters Phuong Thi Doan and Phuong Thi Ngoan and other family members all helped support each other in my absence, and thus helped me too, during my prolonged period of study abroad Thank you all for your support and en-couragement throughout this experience

ade-As a last word, this dissertation could not have been completed without the help

of my classmates and friends, Edward Gotham, Mr Wu, Mike Cuong, and Dew In the five years we studied together I learned so much from you all

Publications

Chapter 2 of this thesis is based on results produced in the following published paper: Phuong Huu Khiem, Yu-Chen Kuo (2019) EC0249: The impact of health insur-ance reforms on children’s educational attainment: Evidence from Vietnam 3rd Inter-national Conference on Econometrics and Statistics, National Chung Hsing University, Taichung, Taiwan

Chapter 4 of this thesis is based on results produced in the following published paper: Phuong Huu Khiem et al (2020) Does tuition fee policy reform encourage poor children’s school enrolment? Evidence from Vietnam Economic Analysis and Policy,

66, 109-124 https://doi.org/10.1016/j.eap.2020.03.001

Summary

In the last decade economic growth among developing nations has been especially rapid However, as countries continue to grow public policies and public policy administration must keep pace with the needs of the society they intend to support National progress requires intensive investigation and accurate identification of policy issues so that policy makers may effectively plan and implement changes

In terms of economic progress, Vietnam is an example of how economic development, poverty reduction and human capital accumulation go hand in hand After determining that human capital and household development were vital to Vietnam’s continued success, policies in these areas began to be proposed and adopted This has led to human capital accumulation, improved household economics, and benefits in health and education sectors

This dissertation seeks to quantify some of these benefits and examine to what degree the success is attributable to policy changes Using a quasi-experimental difference-in-differences approach, with propensity score matching, this manuscript combines three essays that isolate public policy reforms in various sectors and measures the outcomes

The first essay examines the effects of national health insurance reform on children’s educational outcomes Households in the state sector were unaffected before and after the reform, and so children in that group served as a natural control group, whereas children growing up in non-state employed households formed a suitable treatment group Educational outcomes were measured for the three levels of general education: primary, secondary and high school Results showed that the national health insurance reform improved educational outcomes for children in high school, both in terms of enrollment and school completion likelihood Furthermore, it was shown that children from minority groups, females, those in rural areas, and those from poorer families were less likely to derive the same educational outcomes when compared to their counterparts These findings are the first of their kind using the Vietnam household living standard survey data and would be of value to policy makers in countries that plan to adopt a similar health policy

The second essay was extended findings from the first to evaluate the impact of the national health insurance reform in 2005 on the household consumption It exploits the difference between households in the state sector (control group) and households in

the non-state sector (treatment group) Results showed that the national health insurance reform in 2005 has the strong positive impact on the total household consumption, per capital consumption and non-medical expenditure as well with the treatment group, while it has not the impact on the household medical consumption In addition, the ethnic minority households or households living in rural areas or poor households are likely to decrease spending on the goods than that with the households in counterpart While households with spouse(s) in the higher education and occupation skill level or the richer households are likely to increase their consumption than that in comparison with households in counterpart

The third essay examines the influence of the 2010 policy reform on school enrollment rates at the primary, secondary and high school levels The three levels of education in Vietnam were assessed separately by this study It was found that the policy implementation improved enrollment rates at both primary and secondary levels (compulsory in Vietnam), while high school enrollment rates remained unaffected One

of the largest differences identified was for ethnic minority groups and those in regional areas Minority groups preferred to enroll more than their ethnic majority counterparts

at both the secondary and high school levels; however, there is a significant gap among groups, where children from rural areas were overall less likely to enroll than children from urban areas The cause for this may be that the tuition fees and subsidies only covered a small part of the total cost of education expenditure, or it may be part of the opportunity cost equation that older children face when they come from poor backgrounds and have the chance to join the labor force

Key words: Health insurance, Difference-in-differences, Educational attainment,

Household consumption, Tuition fee exemption, Tuition fee policy, School enrollment

Contents

Pages

Publications ii

Summary iii Contents v 1 Introduction 1

1.1 Overview 1

1.2 Background to public policy reform in Vietnam 2

1.2.1 Public policy 2

1.3 Empirical strategy 7

1.3.1 Basic DID setup 9

1.3.2 DID and DID-PSM framework 10

1.4 Organization of the thesis 13

2 Impact of health insurance reform on children’s educational attainment 14

2.1 Introduction 15

2.2 Literature review 20

2.3 Data 22

2.4 Empirical approach 26

2.4.1 DID 26

2.4.2 DID with PSM 27

2.5 Empirical Analysis 27

2.5.1 Effects on child educational attainment – high school level 27

2.5.2 Effects on child educational attainment - primary and secondary level 32

2.5.3 Robustness checks 34

2.5.4 Identification of impact pathways 37

2.6 Conclusion 40

3 Expansion of National Health Insurance eligibility and effects on household consumption 49

3.1 Introduction 50

3.2 Literature review 51

3.3 Data 53

3.4 Estimation model 56

3.5 Results and discussions 57

3.5.1 Total household consumption and per capita consumption 57

3.5.2 Household non-medical consumption 59

3.5.3 Impact of NHI on household medical consumption 61

3.6 Conclusions .63

4 Impact of tuition fee reform on poor children’ school enrollment rates 64

4.1 Introduction 65

4.2 Data 71

4.3 Empirical approach 76

4.3.1 DID 76

4.3.2 DID with PSM 77

4.4 Empirical Results 77

4.4.1 Effects on primary enrollment 77

4.4.2 Effects on secondary enrollment 80

4.4.3 Effects on high school enrollment 82

4.4.4 Further robustness checks 84

4.5 Conclusions 89

5 Conclusions 102

List of Tables

1.1 National health insurance policy change 3

1.2 Tuition fees policy change 4

1.3 Tuition fees policy in brief history 5

1.4 Starting and completion ages of educational institutions in Vietnam 7

2.1 Descriptive Statistics 25

2.2 Probit models for child enrollment at high school level (ages 15-17) 30

2.3 Probit models for child school completion at high school level (ages 16-18) 31

2.4 Probit models of children’s education attainment at primary school level 33

2.5 Probit models of children’s education attainment at secondary school level 34

2.6 Placebo Test (2007 Fake Intervention) 34

2.7 Probit robustness check results 37

2.8 Statistic test comparison between households in the control and treatment groups .39

3.1 Descriptive Statistics 55

3.2 DID results for total household consumption and per capita consumption 58

3.3 DID results on non-medical consumption 60

3.4 DID result on medical consumption 62

4.1 Descriptive Statistics 74

4.2 Probit model results for child enrollment at primary school (ages 6-10) 79

4.3 Probit models for child enrollment at secondary school (ages 11-14) 81

4.4 Probit model results for child enrollment at the high school (ages 15-17) 83

4.5 Descriptive statistics (2002-2008) 85

4.6 DID estimations for school enrollment (years 2002-2008) 85

4.7 Descriptive statistics (2010-2016) 86

4.8 DID estimations for school enrollment (years 2010-2016) 86

4.9 The result of robustness check 88

List of Figures

1.1 Basic DID premise 9

2.1 NHI population coverage (%) 1993-2008 18

2.2 Primary school enrollment (%) (ages 6 to 10) 19

2.3 Secondary school enrollment (%) (ages 11 to 14) 19

2.4 High school enrollment (%) (ages 15 to 17) 20

4.1 Primary school enrollment (%) (ages 6 to 10) 75

4.2 Secondary school enrollment (%) (ages 11 to 14) 75

4.3 High school enrollment (%) (ages 15 to 17) 75

Chapter 1 Introduction

1.1 Overview

Economic growth and international market integration is transforming Vietnam Since 1986, the economy has changed from a centrally planned economy to a market economy with a socialist orientation So far, Vietnam has achieved much in terms of economic development, international relations and human capital accumulation Vietnam’s membership into the Association of Southeast Asian Nations (ASEAN) in

1995 and World Trade Organization (WTO) in 2007 enabled the establishment of many other bilateral and multilateral relations Vietnam has successfully risen from the lowest income country to a middle-income country in the 2010s (World Bank, 2013) The Vietnamese central government has a policy that is geared towards growth, but states that development must also be accompanied by growth in social equity, quality of human capital, and in an environmentally sustainable way In order to achieve these ambitious goals, the central government has assessed several strategies and possible solutions With regard to healthcare, the health insurance system was initially established in 1992 and later reformed in 1998, 2005, and 2009 With regard to educational development, the government enacted policies to increase the enrollment percentages for children, as well as improve educational quality; one such policy was the tuition fee exemption scheme initiated in 1998, revised in 2005, 2007 and 2010 Through this transitional and transformational phase, public administration and public policy have changing demographic and population demands to meet Thus, public policy research plays a crucial role in correctly orienting strategies Identifying policy issues, collecting information and building scientific foundations to support policy formulation (Hashimoto et al., 2006) are essential tools for social development Regrettably, until fairly recently, public policy research in Vietnam was unable to contribute as effectively as it may have done, due to a number of factors There was a lack of coordinated communication between public policy research institutes and policy-making authorities, there was also a lack of coordination between public policy research institutes themselves, combined with staff that had limited skills As a result

of these limitations, and not for want of effort, much of the research done at the existing institutes was highly theoretical in nature and impractical This posed problems of timeliness and accuracy

In transitional societies where inadequate social security and public service systems lead individuals toward managing their own affairs, such as developing insurance systems for themselves, there is a tendency for citizens to focus primarily on day to day business type decisions Concerns over the well-being of the economy generally, or sustainable development are so far removed from low-income individuals on the Maslow pyramid that it hardly bears consideration This dissertation and the new data

it uses provides recent empirical evidence on several fronts, for example: What constraints and socio-economic policies are people currently facing? What effects are these policies actually having?

This dissertation contributes to the discussion regarding effects of public policy reform (health insurance and tuition fees reduction/elimination) on household decision-making Of the three essays contained here, the first and the second essay examine effects of national health insurance policy reform and household decisions Specifically, the first investigates the impact of health insurance reform on children’s educational attainment, whereas the second analyzes the expansion of health insurance eligibility and effects on household consumption The third investigates the impact of tuition fee policy reform on poor children’s school enrollment All sections use data from the Vietnam household living standard survey (VHLSS), however the findings or methodology may also be of benefit to other policy researchers that are looking at enacting similar policies

1.2 Background to Public Policy Reforms in Vietnam

1.2.1 Public policy

1.2.1.1 National health insurance

National health insurance (NHI) has become one of the most important social protection policies of the Vietnamese central government NHI was initially established

in 1992 to begin the process of ensured sustainable development and equality of healthcare for all citizens The NHI (non-universal) was organized, administrated and implemented by the state

Table 1.1 National health insurance policy change

em-4

National Assembly Representatives, People's cil members, preschool teachers, social welfare tar- get groups, dependents of police and armed forces staff

Coun-14

Workers in non-state enterprises more than 1 ployee, cooperatives, other organizations, war veter- ans, the poor

em-46

2014 Law on NHI Dependents of laborers and cooperative members;

1.2.1.2 Education

Educational development has become an increasingly important strategy for the Vietnamese government There have been many policies implemented with the hope of improving children’s educational outcomes and developing human capital While tuition exemption and educational subsidy were determined to be important in each reform, social target groups were addressed piecemeal

1 Vietnamese central government conducted an innovation “Doi moi” and took significant steps

to opening the national economy in 1986 Equitisation of state enterprises started with a pilot program in 1992 which was later extended in 1996 Although the equitisation process had begun

in 1992, it took considerable time for it to gather momentum Around 2,600 firms were equitised

in the first 13 years of the program; of these, around 2,000 were equitised in the period 2000 - 2005) (Sjöholm, 2006) This means that the NHI policy changes in 1992 (which affected non- state enterprises with more than 10 employees) only impacted around 4% of the total population

Date → 1995 1998 2007 2010 Decree → (Decree, 28-CP, April 29th, 1995) (Decree, no: 70/1998/QD-TTg, March 31th, 1998)

(Decree, no:

112/2007/QĐ-TTg, July 20th, 2007)

(Decree, no

49/2010/NĐ-CP, May 14th, 2010)

Primary

(6 - 10)

Treatment

Poor (Select Border Areas) (P135, Phase 2) n/a 50% reduction 100% reduction + subsidy No change

Table 1.2 Tuition fees policy change

Table 1.3 Tuition fees policy in brief history

1995 The policy in this period stipulated that children in war veterans’

households and the people’s armed forces would receive a tuition fee exemption (this included both tuition and the educational contribution payable to schools)

1998 Tuition exemption for primary school pupils (at public education

institutions) was applied to students with parents who reside in mountainous and remote areas and islands; students with disabilities in the family, those having economic difficulties; orphaned students (having neither parent); helpless students; students attending boarding schools for ethnic minorities, and students whose families were classified as “hungry households” according to State regulations In addition, Decree 70/1998 stipulated a 50% reduction of tuition fees for students from poor households, as classified according to State regulation

2007 2007 saw a series of policies in respect of support services designed

to improve people’s lives through provision of legal aid and to raise awareness of laws In particular Program 135, phase 2 made provision for an additional fee reduction (or exemption in some cases) for students in border communities, highland areas, islands and communes with exceptionally difficult socio-economic conditions Any student classified as a member of a poor household under this decree was supported with an additional 70,000 VND/month

2010 Exempted all students categorized as poor, from tuition fees This

demographic often covered, or was specifically designed to cover (i) the target groups above, such as students from households residing in border communities, highland area, islands and communes with exceptionally difficult socio-economic conditions; homeless orphans; students that were children of soldiers serving in the armed forces and war veterans; students of ethnic minority backgrounds at boarding school, and now (ii) children in poor

households nationwide according to a metric calculated by the central government Other children in a category called “near-poor” became entitled to a 50% reduction in their tuition fees, where their maximum household income was equal to 150% of the income of the “poor” category Moreover, to further reduce the financial burden for children in poor households, the government implemented a subsidy to cover study expenses up to 70,000 VND/student/month

The 2010 policy expanded educational opportunities for children from poorer households nationally, using both a tuition fee exemption (100% reduction) and monthly subsidy approach (70,000 VND/student/month, approximately 3.50 USD) This unique policy implementation, and the income and household data made available for the first time through VHLSS survey enable researchers to examine the impact of the legislative change by the creation of separate control and treatment groups to test the effects

Table 1.4 Starting and completion ages of educational institutions in Vietnam

Primary Secondary High school Grade & year 1 2 3 4 5 6 7 8 9 10 11 12

is equivalent to assessing the causal effect of that program on those outcomes

Accordingly, asGertler et al (2016) show, the key to identifying causal impacts of

a program is finding a valid comparison group to estimate a counterfactual case Thus far, there have been a number of impact evaluation options, including randomized

assignment, instrumental variables, regression discontinuity design, differences (DID), and DID with propensity score matching (PSM) In this study, as the public policy reforms (NHI policy and the tuition fee policy) were applied in a unique and specific way, and because suitable data sources were obtained, this provided an opportunity to apply the quasi-experimental DID and DID with PSM (PSM-DID) approach

difference-in-A growing body of literature concerning this approach has shown it to be a reliable non-experimental evaluation method (Villa, 2016) DID estimation has been widely used where panel data or repeated cross-sections are available One key aspect of DID

is that it facilitates the causal inference of an intervention when time-invariant unobserved heterogeneity may otherwise confound a cause-effect analysis (Abadie, 2005; Angrist and Pischke, 2009) Different specifications of the DID model can also account for observed heterogeneity and incorporate other non-experimental evaluation methods into the analysis Furthermore, DID identifies the average treatment effect on the treated group by comparing the difference in post- and pre-program outcomes between two distinct groups: a group that participates in the program (the treated) and

a group that does not participate (the control) DID can be applied efficiently with either panel data or repeated cross-sectional data, as long as the observed outcomes can be classified into treated and control both before and after the program

Despite the availability of other plausible methods based on the existence of observational data for non-experimental causal inference (that is, matching methods, instrumental variables, regression discontinuity, etc.), DID estimation offers an alternative by generating unbiased results while accounting for time-invariant unobserved heterogeneity Four elements are specific to a DID approach (Figure 1.1): the first one is the availability of a treatment group and a control group; the second is the existence of parallel sections in the pretreatment trends; the third is a clear cutoff time, identifying when the treatment begins so there is a well-defined before and after period; the fourth is the assumption that, without treatment, the treated group would show a trend similar to that observed for the control group Thus, DID treatment effects are obtained when panel or repeated cross-section data are available and a treatment has been administered

Figure 1.1 Basic DID premise

1.3.1 Basic DID setup

A basic DID setup involves two groups (D=1: treated units and D=0: control units); two periods (T=0: pre-treatment period and T=1: post-treatment period); and the poten-tial outcomes (Y1i (t): outcome unit i attains in period t, if treated before t, and Y 0i(t):

outcome unit i attains in period t, if not treated before t)

In order to estimate any causal effect in a DID setup, four assumptions must hold: (i) the intervention must be unrelated to any outcome at the baseline (where allocation

of the intervention was not determined by the outcome); (ii) treatment/intervention and control groups have parallel trends in the outcome (see below for further details); (iii) composition of the intervention and comparison groups must be stable, to support use

of the repeated cross-sectional design; and (iv) there must be no spillover effects Among these assumptions, the parallel trend assumption is the most critical of the above four assumptions to ensure internal validity of a DID model, and also the hardest

to fulfill It requires that in the absence of treatment, the difference between the treatment and control group is constant over time Although there is no statistical test for this assumption, visual inspection is useful when you have observations over many time points It has also been proposed that the smaller the time period tested, the more likely the assumption is to hold Violation of the parallel trend assumption will lead to biased estimation of the causal effect

To estimate the DID treatment effect in (Figure 1.1), we rely on linear regression

The subsequent complementary introduction of control variables, or kernel score matching weights, is similarly specified by linear regression In this basic framework, the estimation process follows:

Where outcome(Y)i is the outcome variable for each unit; period(T)i is a binary iable adding a value of 0 into the baseline, and 1 into any follow-up periods; and

var-(D)i

treated is a binary variable indicating the treatment status for each unit

The expected values of DID estimation are obtained from the interaction of the mated coefficients The estimated coefficients have the following interpretation:

esti-0

 : the mean outcome of the control group at the baseline; 0+ : the mean outcome 1

of the control group in the follow-up; 2: the single difference between the treated and the control groups at the baseline; 0 +2: the mean outcome of the treated group at the baseline; 0+ + 1 2+ : the mean outcome of the treated group in the follow-up; 3

3

 : the DID estimated

1.3.2 DID and DID-PSM framework

1.3.2.1 DID

The definition of DID treatment effects as estimated by the diff Stata command is based on the existence of a pair of before-and-after periods, namely, one baseline (t=0) and one follow-up (t=1) The basic DID framework is dependent on the availability of

two groups of units i, including a treated group to which the treatment is delivered

(Z = i 1) and a control group to which the treatment is not delivered(Z = i 0) The treatment indicator in the DID setting requires absence of any intervention in the baseline for either group (Di t,=0 =0 |Z i =1, 0) , and it requires the intervention to be positive for the treated group in the follow-up (Di t, 1= =1|Z i =1) For a given outcome variable, Y it , the population DID treatment effect is given by the difference in the outcome variable for the treated and control units before and after the intervention The single DID setting is given by

in combination with DID can be used Heckman et al (1998) originally proposed and applied this methodology, and it has been replicated by other researchers with valid results, for instance (Kuo and Lin, 2018; Li et al., 2018; Ichino et al., 2017; Becker and Hvide, 2017; Blundell and Dias, 2009; Abadie, 2005; Blundell et al., 2004) PSM-DID uses kernel matching to match the control group and treatment group at a baseline, leveraging similar characteristics within the two groups

A complementary method to the DID treatment effect is the incorporation of kernel propensity-score weights Apart from the inclusion of control variables, observed covariates can be used to estimate the propensity score (the likelihood of being treated) and to calculate kernel weights, again following Heckman et al (1997, 1998) Instead

of accounting for control variables, this method matches treated and control units according to their propensity score Each treated unit is matched to the whole sample

of control units instead of on a limited number of nearest neighbors To begin, one obtains the propensity score (p )i for both groups

Following Heckman et al (1997), the kernel matching is given by the propensity score, given the covariates, which leads to the calculation of the kernel weights,

i k n i

i k n

p p K

h

p p K

To increase the internal validity of the DID estimation, it is possible to restrict (1.4)

to where the common support is the overlapping region for the treated and control

groups This sample of i units can be restricted to the region defined as

( :i p i[max{min(p Z i| i =0)}, min(max(p Z i| i =0)}])

Complementarily, when treated and control units cannot be followed over the line and follow-up periods, the DID treatment effects can be estimated with repeated cross-sections This is very common when a treatment has been administered to certain regional or demographic groups over several cross-sections The kernel propensity-score matching with repeated cross-section DID treatment effects is specified following Blundell and Dias (2009)

Finally, the balancing property of the treated and the control groups can be tested Given the availability of observable covariates, it can be shown that in absence of treat-ment, the outcome variable is orthogonal to the treatment indicator given the set of covariates Furthermore, the balancing property can be tested against the baseline as

1.4 Organization of the Thesis

This research is organized as follows: Chapter 2 examines the impact of NHI reform

on children’s educational attainment Chapter 3 evaluates the effects of NHI expansion

on household consumption Chapter 4 investigates the impact of tuition fee policy reform on poor children’ school enrollment Chapter 5 provides some concluding remarks and public policy recommendations

Chapter 2 Impact of Health Insurance Reform on

Children’s Educational Attainment

Abstract

Research has shown that parental health shocks and child health status each exert measurable effects on child educational attainment, particularly in low-middle income countries In 2005, the Vietnamese government enacted a new health insurance policy increasing the proportion of population covered by health insurance from 22% of total population to approximately 43% Using a quasi-experimental setup and a difference-in-differences (DID) approach, this chapter examines the unintended effects of health insurance reforms on children’s educational outcomes Because households in the state sector were almost unaffected before and after the reform, children in that group served

as a natural control group, whereas children growing up in non-state employed holds formed a treatment group Educational outcomes were measured for three levels

house-of general education: primary, secondary and high school Results showed that the NHI reform improved educational outcomes for children in high school, both in terms of enrollment and school completion likelihood I found a stronger effect on school en-rollment rather than school completion These findings are the first of their kind using the nationally representative data and would be of value to policy makers in countries that plan to adopt a similar health policy

Key words: national health insurance, public policy, difference-in-differences,

educational attainment

2.1 Introduction

Many politicians and researchers promote the idea that educational development is

a key strategy in national civic and health development For instance, Nelson Mandela (2013) writes, “Education is the most powerful weapon which you can use to change the world Education is the key to eliminating gender inequality, to reducing poverty” Determining factors2 that influence children’s educational outcomes is a complex process; recently, studies have provided both empirical evidence and theoretical frameworks regarding the impact of HI policies that are suspected to have positively influenced children’s educational attainment These studies are particularly relevant and popular in developing countries where health care services are less available, while citizen access to services are constrained by low incomes (Chen and Jin, 2012; Cohodes

et al., 2016; Woode, 2017; Mitra et al., 2017)

Confounding matters, HI may have both direct and indirect effects on children's educational outcomes For example, Levine and Schanzenbach (2009) and Alcaraz et

al (2013), showed that insured children achieve higher academic results in school than those that are uninsured, while Yeung et al (2011) showed that an increased participation rate for HI is associated with an increase in average daily school attendance rates Chen and Jin (2012), used a DID approach to study the impact of HI coverage in rural China on children’s education Their paper suggested that children’s enrollment in the new HI program was associated with both better school enrollment and lower child mortality Another possible indirect effect of HI may be the influence

it exerts on children's educational attainment via inter-generational effects within families Woode (2017) studied the impact of parental HI on children’s schooling in

2 In Vietnam there have been studies concerning the factors that affect children’s educational attainment in general, usually looking at school enrollment Key indicators affecting educational outcomes were parental education, family size, household wealth and child age, rural/urban locale, gender, region and ethnicity (Anh et al., 1998; Nguyen, 2006; Mont et al., 2013; Dang and Rogers, 2015; Giang and Cuong, 2017; Dang and Glewwe, 2017) It was also shown that there is an unequal gap in educational outcomes among groups Boys show a higher percentage of enrollment than girls do, and enrollment in urban areas is higher than that of rural areas For practical purposes, the authors will not extensively discuss the many other factors that could potentially affect educational outcomes

Rwanda, analyzing and proposing a theoretical framework, as well as providing empirical evidence of the spillover effects that insured parents have on their children’s schooling Strobl (2017) also conducted empirical research and found that children of households enrolled in HI schemes worked significantly less than those not enrolled, and enjoyed higher educational achievement

Vietnam is a global success story among the low-middle income countries when considering economic growth, human capital development, education and healthcare Among important public policies, NHI is a monumental policy decision for governments to undertake, and it is now a pillar of the social protection policy of the Vietnamese central government The goal of NHI was to increase the proportion of population covered, and produce universal cover for all citizens Between 1992 and

2010, the NHI scheme experienced three major reforms (1998, 2005 and 2009), the biggest one being the second one in 2005 (July 1, 2005: 63/2005/ND-CP) The new NHI policy in 2005 required that people in the non-state sector, including workers in non-state enterprises of more than one employee, economic cooperatives, war veterans and people in poor households, participate in the NHI scheme Employees that worked

in the state sector were covered by NHI before 2005 Figure 2.1 shows that after the implementation of the 2005 NHI policy, the percentage of population covered doubled, from around 22% in 2004, to around 43% in 2006 For this reason, it is expected that NHI reform data, pre-post 2005, may yield a useful natural experiment to examine NHI effects

Existing studies have shown the positive effects of Vietnamese NHI policy on all residents, but especially the vulnerable groups (children, poor and ethnic minorities); however, so far those studies have only examined direct outcomes For example, some examined the effects of NHI on the reduction in financial burden (Wagstaff and Doorslaer, 2003; Sepehri et al., 2006; Axelson et al., 2009; Sepehri et al., 2011; Cuong, 2012; Van Minh et al., 2013), whereas others examined the impact on healthcare utilization and access (Sepehri et al., 2005; Wagstaff, 2007; Wagstaff, 2010; Liu et al., 2012; Cuong, 2012) These studies have only focused on a target group such as poor people, children under 6, or people in rural areas There are limited studies evaluating spillover effects of NHI on other outcomes such as children’s educational attainment and general education levels

Because expansion of health insurance is a popular human capital accumulation

strategy in many developed countries, developing countries are increasingly following suit However, reckless implementation of such policies, without the requisite foreknowledge of indirect effects can be costly in human and economic terms This chapter takes advantage of the natural experiment opportunity, and goes a step beyond the direct effects of healthcare To our knowledge, this is the first study to take nationally representative data and apply the combination of DID and PSM to directly unravel the connection between parental health insurance and educational outcomes for children In contrast to previous research, which often suffered from potential endogeneity problems arising from self-selection into the respective HI programs, the DID/PSM strategy used here provides a solid basis for viewing the causal effects of parental health insurance on children’s schooling.3 In this case, a majority of the data points toward increased child labor and reduced school enrollment as a strategy for households to cope with income shocks and risk (Mitra et al., 2017) This study continues the discussion in this field of investigation and explains how public health policy can provide financial protection as a positive externality, keeping children in school and enabling them to complete their education

Furthermore, studies from other comparable countries have only focused on the impact of health insurance expansion on children’s outcomes in terms of health, labor participation, and schooling The literature showing effects of NHI implementation on school achievement is sparse Papers published thus far are mostly concerned with enrollment and attendance, but stop short of making a conclusive link to achievement One such example is Le (2008), who showed that in mountainous or rural areas, enrollment was lower than in urban areas, and that children were more likely to leave school and return later; or, if suffering some type of economic or health issue, they would delay attendance for longer periods However the approach likely includes some unwanted or unknown degree of bias as school enrollment is the only measurement device As such, this chapter is one of the first studies to evaluate the effects of the 2005

3 Strobl (2017) points out that self-selection of households into the insurance program produces bias and relies on instrumental variable estimation to draw this inference

NHI reform on children’s educational attainment, and contributes to the growing literature in the field of NHI policy

Indicators used for this study are school enrollment and school completion data across three levels of education: primary, secondary and high school Differences between households in the state sector (control group) and non-state sector (treatment group) will be used to perform the tests The VHLSS, trends are presented first as proportions of school enrollment before the NHI reform, and then separated by control and treatment group Figures 2.2, 2.3 and 2.4 describe the trend of school enrollment across three levels of general education in the period 2002-2008, for households in the control and treatment groups Figure 2.2 shows that at the primary school level, enrollment remained fairly constant after 2005 for children in the control group, but showed a small increased in the treated group At the secondary and high school levels, Figures 2.3, 2.4 respectively, show the control group remains stable over the 2005 point;

in contrast, the treatment group shows a modest increase in secondary level education and high school shows a dramatic increase Comparing Figures 2.2, 2.3, and 2.4 shows that NHI implementation has the strongest effect of high school enrollment For this reason, this chapter’s data will focus more on high school age children

Figure 2.1 NHI population coverage (%) 1993-2008

Figure 2.2 Primary school enrollment (%) (ages 6 to 10)

Figure 2.3 Secondary school enrollment (%) (ages 11 to 14)

Figure 2.4 High school enrollment (%) (ages 15 to 17)

2.2 Literature Review

Existing literature on the effects of HI or NHI on children’s educational outcomes in developing countries can be broadly divided into two categories direct and indirect Direct impacts exert influence through expansion HI to children Healthier children lead

to higher school enrollment and attendance, leading to increased effective study time, and thus better academic outcomes (McDougall, 2004; Levine and Schanzenbach, 2009; Contoyannis, 2010; Yeung et al., 2011; Alcaraz et al., 2012; Garcy, 2013; Cohodes et al., 2016; Bortes et al., 2019) Yeung et al (2011) showed that increased participation rates in NHI associates with increased average daily school attendance Assessing long run effects of this, Cohodes et al (2016), examined the effects of public insurance expansion among children in the US in the 1980s and 1990s Their study concluded that expanding HI coverage for low-income children increased the rate of high school and college completion

Other studies focus on the indirect impact of parental insurance status on children’s health outcomes; such studies are particularly common in low-middle income countries For example, Akobirshoev et al (2017) studied the relationship between parental insurance and insured children’s health outcomes and the authors concluded that insured children of uninsured parents have a worse health status and are at higher risk

of asthma, attention-deficit/hyperactivity disorder, developmental delays, learning

disabilities, and mental disabilities when compared to insured children of insured parents Meanwhile, in the absence of NHI support, parental health shocks can have negative effects on children’s schooling (Gertler and Gruber, 2002; Asfaw and Braun, 2004; Sun and Yao, 2010; Bratti and Mendola, 2014; Alam, 2015; Mendolia et al., 2019) The risk of health shock impacts includes additional financial burden, reduction in labor force participation, lowered labor productivity, and increased poverty rates At the same time, one of the most popular coping mechanisms to resolve these issues in developing countries is decreased parental investment for children, which may entail taking children out of school and into the labor market (Wagstaff and Lindelow, 2010; Bratti and Mendola, 2014)

A large swathe of literature already indicates that NHI can help households mitigate catastrophic expenditures due to health shocks, and thus reduce poverty rates (Wagstaff., 2010; Hamid et al., 2011; Aryeetey et al., 2016; Remler et al., 2017; Alam et al., 2017;

Wu et al., 2018; Dou et al., 2018; Mekonen et al., 2018) Mitra et al (2017) reviewed studies from 2000-2014 that applied rigorous evaluation methods on the impact of health insurance on children in low-to-middle income countries They concluded that most studies were in agreement that health insurance provides households some financial protection from catastrophic spending, despite the fact that the results were less clear for the effects of health insurance on health outcomes and health utilization

It is worth nothing that these studies also indicate that NHI reduces the incidence of child labor (workforce participation) due to the reduced likelihood of children leaving school Hence, there is a lower threat to children’s opportunity to study, and they consequently show better academic performance Landmann and Frolich (2015) examined the effect of HI on child labor in Pakistan, and showed that micro-insurance was widely promoted as a tool to reduce vulnerability to shocks, protecting children from child labor Their findings echo the argument that HI for children works as a strategy to reduce school dropout rates and child labor, particularly in developing countries

In the case of Vietnam, Mitra et al (2016) constructed a short panel-dataset from VHLSS and investigated the impact of health shocks on household expenditure Their findings implied there is an intergenerational transmission between health shocks of working-age parents and their children’s education Specifically, having a health shock led to a considerable reduction in education expenditure Mendolia et al (2019) used

the same data but focused on children’s schooling and labor force participation; their results indicated that maternal illness decreased the likelihood of being enrolled in school, and increased the working hours of children aged 11-23 Palmer et al (2015) exploited a regression discontinuity design to evaluate the effects of a public HI reform policy which automatically includes children under age 6; they found that this new policy led to greater utilization of both outpatient and inpatient care

There have also been studies showing evidence of the intergenerational effects of parental HI on children’s education using single period cross-sectional data Woode (2017) highlighted the impact of parental insurance on children’s educational attainment with both a theoretical framework and empirical evidence In the same paper, Woode used data from Rwanda and empirically showed that insured parents were more able to keep their children in school when faced with health shocks, compared to their uninsured counterparts Strobl (2017) used the same data to examine the effects of parental health insurance on child labor and education outcomes and the findings were

in agreement with Woode (2017), showing that parents with health insurance also have children (age 13-18) with higher school attendance rates

HI has been shown to be a key to improving labor productivity and labor force participation This in turn leads to improvements in household incomes and increases

in children’s educational investment Dizioli and Pinheiro (2016) indicated that workers with health coverage missed on average 76.54% fewer workdays than uninsured workers, and HI reduced the probability that a healthy worker gets sick, misses workdays, and increases the probability that a sick worker recovers and returns to work

Su et al (2017) suggested that apart from reducing uncertainty in future spending due

to catastrophic illness, universal HI could also have other positive effects on the labor market, namely, that of increasing rates of self-employment This study is also consistent with a study by Lin et al (2018)

2.3 Data

Data was obtained from the VHLSS, a large and high quality database, with survey information collected by the Vietnam General Statistical Office (GSO), using methods and techniques recommended by the World Bank Collected for the first time in 1992, then again in 1998 and 2002, this survey was thereafter conducted every 2 years reliably The VHLSS includes detailed information on the characteristics of individuals,

households and communities, such as the demographics of household members, ethnicity, area of residence, educational background, employment status, income level, expenditures, housing type, household assets, and utility usage

This study used data from four waves of surveys 2002, 2004, 2006 and 2008, the sample size for the expenditure module was also included In 2002, there were 30,000 households with 132,376 individuals, 2004 had 9,000 households with 40,439 individuals, 2006 had 9,189 households with 39,071 individuals, and the last round in

2008 had 9,189 households surveyed for both income and expenditure, with data on 38,249 individuals

The VHLSS also provides information of the highest completed grade and school enrollment status of the respondents Given this information, school enrollment and school completion data across three levels of general education were constructed using

primary, secondary and high school levels as a dummy variable for empirical analysis

To compare and analyze the impact of NHI reform in 2005 two groups were defined Those where NHI was implemented, the target group (treatment group), and the unaffected group (control group) More specifically, the control group was defined

as households in the state sector, having either a father or mother working in the state sector The treatment group was defined as households in the non-state sector, where households had both father and mother (where families had both parents) or a single parent (in the case of single parent families) working in the private sector The treatment group also contained those with family owned companies, economic collectives, and the private economic sector, foreign investment companies and the self-employed Several sensitivity regressions were conducted by varying the sample size of the treatment group In the first case, only two parent households were used, in the second case single parent and two parent households were combined The sample size for the first case was 44,711 and the second case was 47,657 The tests showed low significance in the first case, and high in the second; however, coefficient signs for both cases were in agreement For the final DID, the combined single and two parent household number was used

The full list of dependent and independent variables used in this paper are shown

in Table 2.1 Table 2.1 shows the descriptive statistics for the household sectors and NHI status With the households in the state sector, a comparison between pre/post 2005

NHI policy indicates that the percentage of enrollment is unchanged at both primary and secondary levels While the proportion of school enrollment in high school slightly decreased over time By contrast, households in the non-state sector saw an increase in rate of enrollment at all levels of compulsory education, with a remarkable degree of statistical significance The highest increase was in children of high school age, showing a 6.7% increase after the policy change

School completion as an indicator of educational attainment also showed an increase in all three educational levels, for both household sectors In the comparison between state and non-state households, the proportion of children completing school from the non-state sector increased

Table 2.1 also describes the children’s characteristic variables, parental features and household control variables, including household size, the percentage of children under 15 years old (dependents), household income, and type of house In early test regressions this chapter attempted to use number of children as a proportion of household size as an explanatory variable, but the results produced inconsistent coefficient signs When the factor was converted to a percentage: (number of children

<15 / household size)*100, the coefficient signs for the different models were in agreement Using this method, it was then possible to use number of children in relation

to household size (ratio) expressed as a percentage as an explanatory variable

For consideration of household net income, this chapter used the World Bank consumer price index (CPI) for 2002, 2004, 2006 and 2008 to adjust for inflation, all calculations were made in 2002 VND Unsurprisingly, the family income for households in the state sector is higher than that of households in the non-state sector, which is likely due to the higher education level of the parents in the state sector This

is a similar finding to other countries where higher levels of educational attainment lead

to improved employment opportunities

Table 2.1 Descriptive Statistics

Household sector

Variables

Pre-HI (2002- 2004)

Post-HI (2006- 2008)

DIFF a Pre-HI

2004)

(2002-Post-HI (2006- 2008)

11-14) .975 .975 .000 (.06) .882 .907 .025** (4.72) School enrollment (High school ages

15-17) .891 .879 -.011 (-.79) .5903 .657 .067*** (7.27)

School completion (Primary ages

7-11) .884 .930 .045** (3.23) .736 .845 .110*** (14.82) School completion (Secondary ages

12-15) .788 .897 .109*** (6.92) .503 .731 .229*** (28.34) School completion (High school ages

Mother's years of schooling 9.339 9.362 5.921 6.188

Note: Mean difference (DIFF) test was performed on pre/post NHI period by household sector,

t-statistics are in parentheses * p<0.05, ** p<0.01, *** p<0.001

2.4 Empirical Approach

2.4.1 DID

The 2005 policy change on NHI allows researchers to target households in state and non-state sectors The switch from non-compulsory to compulsory for non-state households permits the use of a DID approach to identify the effects of the change Following a similar methodology to previous studies, Chou and Staiger (2001), and Kuo and Lin (2018), it is possible to estimate the effect the NHI reform using the following probit equation:

Pr(Yit = =  1)   + HIit +  Hhit+  HI Hhit it +  Chit+  Fmit (2.1) where Yit represents enrollment or completion, and presents a dummy variable for enrollment and completion in different levels of primary, secondary and high school levels In this chapter, we use both of school enrollment and completion as indicators

to evaluate the effects Although this provides more statistical power to the estimation, there may be some differences between the two indicators School enrollment assesses the schooling status of children (in a yes/no format) School completion, however, measures the precise achievement level attained including age and grade, as seen in Table 1.3

HI is a dummy variable that identifies the period after implementation of health insurance After 2005, HI equals 1, and 0 otherwise Hh represents the household’s status as state or non-state sector HIHh is the interaction term between HI and Hh, which will yield the DID estimator Ch represents a set of children’s personal characteristics including age, ethnicity, gender, urban or rural location Fm is a vector

value calculated using a mixture of family characteristics such as: parent age(s), parent(s) years of schooling, household income, type of house, household size, poverty (yes/no), percentage of children under 15 years old, and location, which captures an

overall wealth status

The selection of these variables resembles personal-level determinants of school enrollment used by many researchers in previous studies, such as Dang (2017) and Mendolia (2019) In this model, 3 represents the DID estimator of the impact of NHI implementation on educational outcomes

2.5 Empirical Analysis

Determining the impact of NHI implementation on children’s educational attainment is achieved by considering households with at least one parent that works in the state sector as the control group, and households whose parents work in the non-state sector as the treatment group The probit model was used to estimate separate coefficients for school enrollment and school completion at the high school level Analysis was then extended to children at primary and secondary school levels

2.5.1 Effects on child educational attainment – high school level

Table 2.2 presents the results for several specifications of the coefficients for the interaction and determinants of school enrollment Examining differences between Figures 2.2, 2.3, and 2.4, these estimates allow us to control for observable characteristics that differ between the state/non-state household’s sectors For reading convenience, the coefficients from the probit analysis were transformed into the marginal probabilities implied by those coefficients

There is strong evidence that children in households from the non-state sector were more likely to enroll in high school after NHI implementation The coefficient of the interaction term is statistically significant for all four models The first model (column 1) shows the child characteristics, parent characteristics and household characteristics The result indicates that an increase in the child’s age leads to a decline in the probability of enrollment, with female children being less likely to enroll than boys This finding is consistent across all models As seen in Table 2.2, in almost all models, children in the ethnic minority category live in mountainous regions Vietnam has 54 minority ethnic groups, the Kinh (Viet) representing the majority, which is 86% of the population (General Statistical Office, 2008) The minority groups live in difficult conditions, and account for the highest rate of poverty, and show a positive effect in enrollment

In contrast, rural effects were negative on children’s schooling This is consistent across primary and secondary levels In columns 2 and 3, we added more independent variables to control for the household characteristics (household size, percentage of children under 15, and poverty status) The result shows that maternal age has a negative effect on high school enrollment, with a weak but statistically significant result Paternal age is a positive effect on child schooling, with a high level of statistical significance Moreover, both maternal and paternal education show a positive impact on child education Household size and number of children effects are both negative Model 4 used more variables to capture locale The result showed that the probability of high school enrollment was higher for children living in the north and central regions of Vietnam, including the Red river delta, North West, North East, North Central Coast and South-Central Coast, then for regions in the south of Vietnam4

Table 2.3, adds another indicator, school completion, to measure the impact of NHI reform The result shows strong evidence that school completion receives a positive

4 Vietnam has 8 distinct regions: the Red River Delta (encompassing Hanoi, the capital), the Northeast, the Northwest, the North Central, the South Central, the Central Highlands, the Southeast (encompassing Ho Chi Minh City, the largest city), and the Mekong River Delta However, economic development varies widely among these regions

effect from NHI implementation The coefficient from the DID is positive for school completion, and while it is only weakly statistically significant, the result is reproduced across all four models Using a similar strategy with regression on the school enrollment indicator, the first column captures only child characteristics, parent characteristics and family income In the next model (columns 2 to 4) I added more variables to control for household wealth, poverty status, family size, percentage of children age under 15 years old, and regions Almost results are consistent with the results of school enrollment Comparing school enrollment and school completion, the empirical results from this paper show that HI exercises a larger DID effect on school enrollment rates than school completion This may be explained by the education system in Vietnam If children do not meet a certain academic standard for a school year, they will be required to repeat

a grade That is, they retain their “in school” status Le (2008) already showed that in remote and mountainous areas, children were more likely to leave school and return later due to economic or health issues, which would lead to some distortion of the observations and result in an overstatement of HI effects when only using school enrollment as an indicator for educational outcomes One other notable finding was that children from minority families were more likely to enroll (positive coefficient), implying that they benefitted from the NHI reform; however, there was a negative effect (coefficient) on school completion This was consistent with findings by Le and Homel (2015), where children from minority ethnic groups were more likely to be employed part time, and thus achieve lower academic results than children from the majority ethnic group This is also consistent with results from Bui et al (2014), where ethnic minorities were found to be more vulnerable to natural disasters and economic shocks than the ethnic majority Furthermore, Le (2008) showed that children in ethnic minority groups would frequently enroll at school then leave before completion These factors lead to repeated grade failures and lower achievements in education overall for ethnic minority children