Impact of health insurance in Vietnam on healthcare utilization, self-reported health, and financial choices.Impact of health insurance in Vietnam on healthcare utilization, self-reported health, and financial choices.Impact of health insurance in Vietnam on healthcare utilization, self-reported health, and financial choices.Impact of health insurance in Vietnam on healthcare utilization, self-reported health, and financial choices.Impact of health insurance in Vietnam on healthcare utilization, self-reported health, and financial choices.Impact of health insurance in Vietnam on healthcare utilization, self-reported health, and financial choices.Impact of health insurance in Vietnam on healthcare utilization, self-reported health, and financial choices.Impact of health insurance in Vietnam on healthcare utilization, self-reported health, and financial choices.Impact of health insurance in Vietnam on healthcare utilization, self-reported health, and financial choices.Impact of health insurance in Vietnam on healthcare utilization, self-reported health, and financial choices.
INTRODUCTION
Development of health insurance in Vietnam
Vietnamese health insurance has been in place for almost 29 years It had been piloted in various areas before 1992, including Hai Phong, Quang Tri, and Vinh Phu The government issued Decree 299/HBT on August 15, 1992, and it was in force from 1992 to 1998 According to the decree, required enrollment groups include public servants, retirees, and workers in businesses with more than 10 employees, while optional registration groups comprise the others (The Government of Vietnam, 1992) Depending on the type of insured, the premiums are fixed at 3% of total salaries, minimum salaries, or retirement pensions The health insurance was administered by the Ministry of Health, and the Vietnam Social Security Agency (VSS) was responsible for its implementation (The Government of Vietnam, 2002). The ministry of health is responsible for formulating health insurance policies, including determining premiums, establishing the benefit package, and determining reimbursement rates and co-payments (Q N Le et al., 2020).
In 1998, cadres and commune-level government officials were added to the list of those eligible for the mandatory system, and co-payments were introduced first time(The Government of Vietnam, 1998) The health insurance fund pays 80% of medical examination and treatment expenses, while the patient pays the remaining20% to the healthcare facility, except for veterans A co-payment ceiling of six months' base salary is in effect.
There were several modifications to the health insurance policy in 2005. Specifically, workers with minimum three-month contracts at businesses with fewer than 10 employees were added to the required categories (The Government of Vietnam, 2005b) Children under the age of six qualified as noncontributory health insurance recipients The overall co-payment ceiling was eliminated; however, it remained for outpatient treatments (only in voluntary schemes) and a list of costly services (both schemes) (Q N Le et al., 2020).
In 2008, the National assembly passed law No 25/2008/QH12 regarding health insurance with a range of changes To be specific, 25 categories of individuals were eligible to enroll in the health insurance program, 20 of which were required, and the other five were optional Beginning in 2010, the premium rate rose from 3% to 4.5% of monthly salaries, base salary, retirement pensions, or unemployment benefits, depending on the kind of insured All co-payment ceilings were lifted, and three levels of co-payments were established: 0% (children under the age of six, persons who have lost working capacity, and the unemployed); 5% (poor people, veterans, and those receiving social allowance); and 20% (all the others) (The National Assembly of Vietnam, 2008) When it comes to health insurance for students, prior to 2010, students were grouped into voluntary categories; however, beginning in January of that year, membership in health insurance plans for students became required As of January 1, 2011, adults aged 80 or older will get free health insurance under the program for senior healthcare (The National Assembly of Vietnam, 2008, 2009).
The National assembly passed amendments to the health insurance law in 2014, which went into effect on January 1, 2015 The law reduces the number of enrollment groups from 25 to 5, eliminating voluntary schemes (Q N Le et al.,
2020) In particular, the law contains a number of new regulations that were not included in the prior version One of the significant new regulations is the requirement for treatment cost reimbursement if the insured visits a non-registered hospital When a cardholder
6 is treated at a health facility other than his registered hospital, the insurance fund will cover 40% of inpatient treatment expenditures at central hospitals, 60% of inpatient treatment expenditures at provincial hospitals, and total medical examination and treatment expenditures beginning January 1, 2016 (The National Assembly of Vietnam, 2014) Another new regulation is that people in a household will get reduced prices if they buy health insurance in bulk That is, for household- based enrollment groups, the premium for the first member in a family is 4.5% of the base salary, while the second, third, and fourth pay 70%, 60%, and 50% of the first's premium, respectively, and the fifth, and so forth, only pay 40% of the first's rate (The Government of Vietnam, 2014) With regard to co-payments, a 5% co- payment has been abolished for the poor and veterans A co-payment ceiling has been reintroduced for individuals who have been members for at least five consecutive years (Meiqari et al., 2019).
Table 1.1: Health insurance reform in Vietnam
Key years Health insurance reform in Vietnam
Health insurance was piloted in three provinces in 1990 With the implementation of Decree 299/HBT, voluntary and mandatory health insurance schemes were introduced nationally in 1992.
- In 1998, the enrollment list was expanded to include commune- level cadres and civil servants The premiums are fixed at 3% of total salaries, minimum salaries, or retirement pensions.
- Introduction of a co-payment mechanism: patients pay for 20% of medical expenses, and the health insurance fund pays for the remaining 80%.
Key years Health insurance reform in Vietnam
2005 - Enrollment was made compulsory for workers at businesses with fewer than ten employees.
- All children under six years are covered by free health insurance.
2009 - The law on health insurance was effective and classified eligible people into 20 compulsory groups and five voluntary groups.
- Co-payments were divided into three categories (0%, 5%, and 20%). All co-payment ceilings were removed.
2010 - Enrollment was compulsory for students; however, they had their premium reduced.
- The premium rate rose from 3% to 4.5% of monthly salaries, base salary, retirement pensions, or unemployment benefits.
2011 People aged 80 or over were eligible for free insurance.
2015 The amended law on health insurance abolished voluntary enrollment and introduced new regulations that were beneficial to cardholders A co-payment ceiling was brought back for people who have been members for at least five years in a row.
Sources: Own construction based on The Government of Vietnam (1992, 1998,2005a, 2005b, 2014), and The National Assembly of Vietnam (2008, 2014)
Healthcare scheme in Vietnam
Prior to Đổi Mới (reform), Vietnam's healthcare sector had been centralized, and the state had been the only supplier of services Along with economic reform beginning in the late 1980s, Vietnam's healthcare system also underwent a transformation beginning in 1989, shifting from a completely public service scheme to a hybrid public-private provider scheme (D.-C Le et al., 2010) Including the private health sector and implementing user-fee for services resulted in more options and more opportunities for individuals to get better healthcare (D.-C Le et al., 2010) Over the last several decades, there has thus been a remarkable improvement in people's health, and the world community has recognized Vietnam as one of the top 10 high-
8 performing nations in its attainment of the Millennium Development Goals for health (Teo & Huong, 2020).
The administrative hierarchy of the healthcare system is comprised of four levels that correspond to the levels of state administration (see Figure 1.1): central, province, district, and commune (Teo & Huong, 2020) The commune health center is the first point of care, typically consisting of a doctor or assistant doctor, a midwife, nurses, an assistant pharmacist, and a network of village health workers, serving a population of 5,000–20,000 (Meiqari et al., 2019) Commune health centers offered a variety of basic services, including care for mothers and children, family planning, treatment for acute respiratory infections, vaccination, and care for common diseases (D.-C Le et al., 2010) The commune health center is responsible to the district health bureau and the Commune People's Committee for the protection and promotion of local healthcare, and it receives technical assistance from the district hospitals (Van Minh et al., 2014) District and provincial health facilities are administered by the Ministry of Health and are responsible for implementing and developing healthcare services at their respective levels (D.-C Le et al., 2010) At these levels, the people's committee is responsible for distributing financial and personnel resources, while the province or district health bureau is in charge of professional competence under the Ministry of Health's supervision (D.-
C Le et al., 2010) The Ministry of Health is the central government agency in charge of government healthcare protection and promotion, which includes preventive medicine, curative care, rehabilitation, traditional medicine, prophylactic and treatment drugs, cosmetics, food safety and hygiene, medical equipment oversight, and management of public services under ministry control (Van Minh et al., 2014).
Figure 1.1 Organisational chart of the Vietnamese healthcare system
Source: Own construction based on Meiqari et al (2019)
Payments in Vietnam's health insurance are made on a tripartite basis (see Figure 1.2) In order to participate in health insurance, both public and private providers (hospitals and health centers) are required to have a contract with the VSS The insured pay VSS premiums, but they also pay co-payments when they visit a healthcare facility for a medical examination and treatment Furthermore, if they utilize medical services not covered by health insurance, they have to pay a user-fee for service (Ha et al., 2021) The contracted providers are responsible for delivering
Government Central government Ministry of
Central hospitals (General and Specialized)
Provincial Hospitals (General and Specialized)
District Hospitals, District Health Centers
The insured Co-payments, User-fee for service
10 services to insured individuals and claiming reimbursement from the VSS (Ha et al., 2021).
There are currently three kinds of provider reimbursement mechanisms that can be employed, including fee-for-service, capitation, and payment by diagnosis-related groups (DRGs) (The National Assembly of Vietnam, 2014) Under the fee-for- service model, the providers are reimbursed by the VSS for each service they provide Under capitation, the provider gets paid in advance at a predetermined fixed rate to deliver a defined set of services to each person who enrolls with the provider for a certain period of time (Ha et al., 2021; H T H Nguyen et al., 2017). With payment by DRGs, the provider is paid at a predetermined rate per discharge, depending on diagnosis, treatment, and type of discharge (Ha et al., 2021).
Figure 1.2: Payment mechanism in health insurance Source: Own construction based on Ha et al (2021)
In 1995, the social health insurance agency formally implemented fee-for-service payment as a payment method after the legalization of user fees at government health institutions (Tien et al., 2011) Despite being considered unsustainable, fee- for- service payment is a prevalent reimbursement method from the health insurance fund to Vietnam's healthcare providers (Ha et al., 2021; Q N Le et al., 2020).According to the Ministry of Health & Health Partnership Group (2011), the fee- for-service
11 payment mechanism contributes to the overprovision of health services (primarily due to the service provider) and results in the waste of resources, the escalation of medical service prices and higher medical expenditures for the whole society In response to the fee-for-service payment mechanism's limitations, the capitation payment model was introduced to achieve effective cost management, emphasizing better healthcare at reduced prices (JICA & KRI International Corp, 2017). Capitation was piloted for the first time in 2005, and starting in 2010, many district hospitals switched from fee-for-service to capitation for insured patients (H T H. Nguyen et al., 2017) As of now, diagnosis-related payment is still in the pilot stage.
This thesis seeks to provide new understanding and empirical evidence on the relationship between health insurance and healthcare behavior in Vietnam The specific objectives are as follows:
1 To assess the degree to which the health insurance program for the elderly facilitates healthcare utilization and provides financial protection to covered individuals.
2 To explore the impacts of having health insurance on households' choices of financial services such as private insurance, savings, investments, and credit.
3 To examine the impacts of health insurance co-payments in Vietnam on the self-reported health of those covered by the plan.
This thesis is expected to address gaps in the literature on health economics by providing methodologically sound evidence on the effects of health insurance on the elderly, adding information on the effects of health insurance on households' financial decisions, and creating country-specific evidence on the effects of a health insurance cost-sharing policy on self-reported health It is expected that the findings obtained
12 from this thesis will add to the current debate over the impact and effectiveness of health insurance policies in Vietnam and the contexts of other countries.
This thesis seeks to investigate the effect of health insurance in Vietnam on measures of financial protection, healthcare usage, households' financial choices, and health Based on the research objectives, the specific research questions are as follows:
1 What are the impacts of health insurance for the elderly on the probability of an outpatient visit, the probability of an inpatient visit, the number of outpatient visits, the number of inpatient visits, the expenditures per outpatient visit, and the expenditures per inpatient visit?
2 What impacts does health insurance have on households' choices of private insurance, savings, investments, and credit?
3 What impact do co-payments in health insurance have on reported health among the insured?
This thesis is to focus on the impacts of health insurance on certain healthcare behaviors, such as healthcare utilization, out-of-pocket expenditures, household financial decisions, and health status The selection of secondary data for analysis determines the scope of the present thesis Specifically, for the research on the effects of health insurance on healthcare use and out-of-pocket expenditures, the current study centers only on the behaviors of elderly adults aged around 80 years in rural regions of 63 provinces and cities in Vietnam The time frame for the analysis is from 2012 to 2018 The scope of this study on the effects of health insurance on household financial choices is restricted to rural households living in the provinces of Ha Tinh, Thua Thien Hue, and Dak Lak in 2013, 2016, and 2017 The boundary of the study
13 on the effect of the co-payment program on health status is limited to rural individuals in the provinces of Ha Tinh, Thua Thien Hue, and Dak Lak in 2017.
The following are the boundaries of this thesis on the concepts of health insurance and healthcare behaviors: healthcare utilization in this research is confined to two proxied variables—the probability of an inpatient or outpatient visit and the number of inpatient or outpatient visits Out-of-pocket expenditures are defined as expenditures per inpatient or outpatient visit The monetary values of health insurance and financial services are beyond the scope of this thesis Rather, it solely focuses on the choices individuals or households make about health insurance and financial services; hence, they are proxied by binary variables in the econometric models In the current thesis, health status is just based on what people say about their own health (self-reported health).
The thesis is organized in an essay-based format, with three essays addressing different health insurance-related impacts in Vietnam This thesis is divided into six major chapters, which are as follows:
This chapter begins with an introduction, followed by a review of Vietnam's health insurance and healthcare system and a brief discussion of the research objectives and questions.
THEORY AND LITERATURE REVIEW
Model of demand for health
Grossman was among the first to study the model of demand for health (Grossman, 1972a, 1972b, 2000) The Grossman model of demand for health has been generally applied by many researchers in the health economic field (Mwabu, 2007) In this model, demand for healthcare is derived from the demand for health investment Put it differently, individuals need good health; accordingly, in seeking desired health status, they demand medical care inputs to produce it Health is a consumption good because it reduces the number of sick days (thereby increasing utility) Additionally, it is also an investment good due to the fact that it increases the number of healthy days available for market and non-market activities (Grossman, 1972a) InGrossman’s model, people are endowed with an initial stock of health, which depreciates over the years, but can be increased by investment Individuals invest in health by consuming healthcare and combining exercise, diet, and time These investments help maintain or improve people’s health stocks, which in turn provide them with healthy days (Folland et al., 2013).
Individuals with constrained budgets and personal preferences may utilize a utility- maximizing framework to determine the optimal mix of healthcare and other goods. Mwabu (2008) presented the unified model of healthcare demand created by Grossman (1972a) and Rosenzweig & Schultz (1982) Specifically, each individual has a utility function determined by health-neutral goods, health-related goods, and health status.
Where X is a vector of health-neutral goods that provide utility but do not have an obvious impact on a person's health, such as clothes or bus transportation Y denotes a vector of health-related goods that both influence health and directly contribute to utility, such as exercise or smoking H denotes an individual's health status.
Let the function below describe the production of health (H) by an individual:
Where Z denotes a vector of health investment goods (e.g., medical care) that enter the utility function only via their impact on H μ) denotes a random component of health that depends on genetics and the environment The budget constraint for an individual is:
Where I represents exogenous money income and P M , P Y , and P Z represent the prices of health-neutral goods, health-related goods, and health investment goods, respectively Therefore, the maximizing problem is written as follows:
Upon solving the maximization problem, the demand functions have the following general forms:
Equation (2.6) is the direct demand function for healthcare, stating that the demand for healthcare services is dependent on prices, P, income, I, and μ) This demand function is a quantity-price relationship that illustrates how the consumption of a certain healthcare service is affected by its own price P Z , all other variables being held constant (Mwabu, 2008).
As the rule of the downward-sloping demand curve is the most fundamental law in economics, the quantity of health demanded should be inversely associated with its
“shadow price” (Grossman, 1972a) Grossman pointed out that the shadow price of health is affected by a wide range of factors, including the cost of medical services.
He asserted that, under certain conditions, a reduction in the price of a health input decreases the shadow price of health and increases the amount of health demanded (Grossman, 1972a) Lowering the cost of health inputs such as medical care use is expected to raise demand for healthcare (L Chen et al., 2007).
Given Grossman's framework, this thesis predicts that the relative decrease in out- of- pocket prices for healthcare services after the implementation of Vietnam's free health insurance program for the elderly would result in an increase in healthcare utilization However, a reduction in out-of-pocket expenditure is not always guaranteed Al- Hanawi et al (2021) argued that the sign of the impact of health insurance coverage on out-of-pocket expenditure depends on the generosity of the package Individuals covered under a full coverage plan are likely to reduce out-of- pocket expenditures and may have little motivation to utilize the savings in higher levels of healthcare
(Al-Hanawi et al., 2021) A partial coverage plan makes the insured incur out-of- pocket expenditures If the savings from utilizing health insurance exceed these out- of-pocket expenditures, overall out-of-pocket expenditures will decline If the savings solely balance out-of-pocket expenditures, there is no change in the overall amount of out-of-pocket expenditures However, if the savings are smaller than these out-of- pocket expenditures, the overall out-of-pocket costs will rise Given that the health insurance program for the elderly in Vietnam offers full coverage for both outpatient and inpatient care without any co-payment, it is predicted that the insured experience fewer out-of-pocket expenditures than the uninsured among those who use services.
The behavioral model of health services utilization
This model was originally developed by Ronald M Andersen in the 1960s to explain why families utilize healthcare, then revised by himself, and it has gone through four phases (R M Andersen, 1995) In this model, Andersen revealed that the utilization of health services is a function of three groups of factors.
- Predisposing factors imply the characteristics that make some people have the propensity to utilize more health services than others These factors include education, occupation, ethnicity, social networks, social interactions, culture, attitudes, values, the knowledge that people have concerning and towards the healthcare system, age, and gender (R Andersen & Newman, 1973).
- Enabling factors essentially refer to conditions that make healthcare services available to predisposed individuals These conditions can be measured by personal/ family resources which consist of access to health resources, income, level of health insurance coverage, and a regular source of care In addition to personal/family resources, enabling attributes of a community can exert influence on the utilization of healthcare services These characteristics are the availability of health personnel and facilities Besides, the rural-urban nature of the community in which individuals live may affect the utilization (R Andersen & Newman, 1973).
- Need factors serve as the most immediate cause of health service use In the presence of predisposing and enabling conditions, people must perceive illness or the probability of its occurrence for the use of health services to occur Need factors can be measured by the number of disability days or a self-report of the general state of health (R Andersen & Newman, 1973).
Based on Andersen’s behavioral model (see Figure 2.1), health insurance is considered an enabling factor for healthcare utilization The model thus gives a theoretical look at how the health insurance program for the elderly affects healthcare utilization.
Education, occupation, ethnicity, social networks, social interactions, culture, attitudes, values, knowledge, age, and gender
Regular source of care Community Rurality The availability of health personnel and facilities
Perceived Need Chronic health conditions Overall physical health status
Utilization Inpatient visit Outpatient visit
Figure 2.1: Andersen’s Behavioral Model of Healthcare Utilization
Source: Own construction based on Simmons et al (2008)
Moral hazard
When studying the impact of health insurance, it is essential to briefly review the theory of moral hazard It basically refers to phenomena in which more informed individuals alter their behaviors in such a way that causes the cost to less informed individuals The term “Moral hazard” was first used in the health economic field by Kenneth J Arrow in an article published by The American Economic Review (Arrow, 2004) He asserted that physicians work as an agent of insurance companies And, the physicians are not under adequate control; they may therefore have the incentive to prescribe more expensive medication that induces financial burden to the companies In general, Arrow (2004) posited that health insurance coverage leads to cost rises for insurance companies Although Arrow (2004) did not directly refer to moral hazard as a problem of morality, in response to it, Pauly argued that “the response of seeking more medical care with insurance than in its absence is a result not of moral perfidy, but rational economic behavior” (Pauly,
1968) From Pauly’s perspective, the increase in quantity demanded for medication when people have insurance is similar to the situation in which the quantity demanded at zero price is higher than that at a positive price (Pauly, 1968) Arrow and Pauly are among the first to lay the foundations for the development of the theory of moral hazard in the health economic field.
Ehrlich and Becker (1972) studied in depth to develop a theory that emphasized the interactions between insurance and two preventive activities called self-insurance and self-protection In their view, self-insurance is the action to reduce the size of a loss, and self-protection is essentially activities to decrease the probability of a loss And they concluded that insurance and self-insurance are substitutes; insurance and self-protection are complements (Ehrlich & Becker,
1972) From the description of these two preventive behaviors, moral hazard can be divided into two distinct types, which are called ex-ante moral hazard and ex-post moral hazard (Chun-Wei Lin, 2012) Essentially, both types of moral hazard behaviors occur following individuals purchasing insurance Ex-ante moral hazard refers to the phenomena prior to the advent of illness in which insured individuals engage in risky health behaviors, increasing the probability of a loss Ex-post moral hazard is related to the increased consumption of healthcare services once an event of illness has occurred (Jowett et al., 2004).
Based on moral hazard theory, the free health insurance program for the elderly in Vietnam is predicted to cause both ex-ante and ex-post moral hazards.After obtaining free health insurance, individuals tend to invest less in preventative measures such as exercise and a nutritious diet This occurrence is referred to as an ex-ante moral hazard And after being ill, people are motivated to increase their healthcare consumption This is called ex-post moral hazard, and the goal of this thesis is to find empirical evidence of it.
Moral hazard and Cost-sharing
According to the conventional moral hazard insurance theory, additional healthcare expenditures in the presence of health insurance are inefficient (Feldstein, 1973; Pauly, 1968) The major reason for welfare losses is that the cost of producing care (reflected in the high market price) exceeds the actual value of care to consumers (reflected in the low insurance price) (Nyman, 2004) To reduce the social welfare losses due to moral hazards in healthcare, health economic theories justify the use of cost-sharing (e.g., a deductible, co-payment, or co- insurance) as a policy tool to limit the utilization of healthcare services (Folland et al., 2013).
Q C Figure 2.2 Conventional ex-post moral hazard effect
Source: Own construction based on Nyman (2007), and Chen (2016))
Figure 2.2 depicts a graphic representation of the conventional ex-post moral hazard The price of healthcare is initially established at P1 Without insurance, Q1 is the optimal consumption level for individuals With full health insurance, medical treatment costs fall to zero, and customers consume Q2 The area ABQ2 measures the deadweight loss due to moral hazard With co-payments, the cost of healthcare at PC would be more than zero but lower than at P1 With PC, overconsumption falls from Q2 to QC In the healthcare market, doctors and patients work together to decide which treatments will be undertaken, but doctors have a better understanding of the potential outcomes of such treatments than patients do (C Chen, 2016) In this context, doctors have some leeway in adjusting the number of health treatments provided; the moral hazard impact may be mitigated most effectively by adjusting the quantity closest to Q1.
However, Nyman (2004) presented a new theory claiming that much of the moral hazard impact on healthcare utilization is efficient According to the new theory, health insurance allows an economy-wide redistribution of income from the healthy
0 Q 1 to the sick He believes that the greater consumption of healthcare caused by insurance price reductions should be seen as a welfare-gain From his perspective, health insurance is a mechanism for redistributing income from the healthy to the sick, who would not otherwise be able to afford the medical treatments essential for sustaining their health (Nyman, 2004) Cost-sharing, he believes, should not be extended to those with severe illnesses since it prevents them from receiving necessary care and reduces the benefits gained from health insurance (Nyman, 2004).
When it comes to the moral hazard consequences, Mendoza (2016) distinguishes between those that reduce welfare (or are undesirable) and those that increase welfare (or are desirable) Undesirable moral hazard results in welfare losses for society, whereas desirable moral hazard increases welfare Moral hazard arising from non- preventive, cosmetic, or habitual needs (e.g., cosmetic surgery, prescription eyeglasses, treatments for hair loss and sexual impotence, visits to the doctor's office to purchase beauty products), which is considered undesirable, should be distinguished from that arising from preventive care and treatment of life- threatening and other serious illnesses (e.g., heart bypass operations, cancer treatment, organ transplantation, trauma), which is considered desirable (Mendoza, 2016).
Following Folland et al (2013), the co-payment policy in Vietnam contributes to mitigating moral hazard effects And, based on the distinction between desirable and undesirable moral hazards (Mendoza, 2016), co-payments reduce both essential and non-essential care This thesis concentrates on the impact of co-payments on lowering essential care, which in turn results in worse health outcomes.
Literature relevant to the heterogeneous impact of health insurance
Grossman's healthcare demand model, Andersen's behavioral model, and the theory of moral hazard predict that expanding health insurance coverage would increase healthcare use However, other theories provide important insights into the fact that the effects of health insurance will not be consistent for a number of reasons. Specifically, health insurance's impact on healthcare utilization is conditional on factors including waiting times and reimbursement strategies.
Wait times for medical treatment have been the subject of literature Patients' frustration with long wait times has been well-documented, and the issue seems to be a consistent and vital contributor to their dissatisfaction (J Sun et al., 2017). Lindsay & Feigenbaum (1984) constructed a model in which the waiting time serves as a rationing mechanism Waiting time is important because the present value of the therapy diminishes the longer it is delayed Due to the high costs associated with joining the queue, some individuals are discouraged from seeking treatment because of the long wait times (Lindsay & Feigenbaum, 1984) Hoel & Sổther (2003) proved that wait times in the public healthcare system induce patients with high waiting costs to seek private care.
The aforementioned healthcare waiting time literature probably offers pertinent arguments as to why the waiting time could lessen the desire for health treatment. Accordingly, it argues that if hospital wait times are sufficiently lengthy, the effect of health insurance on healthcare use may be nullified.
Principal-agent theory and reimbursement methods
In economics, a principal-agent relationship happens when one party (the principal) hires another party (the agent) to do certain tasks on its behalf and gives the agent the power to make decisions (Pontes, 1995; Stephen A., 1973) This dependence on the agent suggests an information asymmetry in which the agent knows more than the principal (Smith et al., 1997) Since the two parties may have different goals, it is possible that the agent acts in his own self-interest rather than the principal's (X.Liu, 2013).
Principal-agent theory can be a helpful framework for interpreting the behaviors of healthcare providers under the Vietnamese health insurance reimbursement scheme.
As mentioned earlier, health insurance reimbursement in Vietnam is typically made on a tripartite basis In this scheme, the VSS is in charge of supervising and reimbursing healthcare facilities for the cost of treating insured patients; hence, it serves as the "principal." Healthcare facilities take on the role of the "agent" since they are the ones responsible for delivering healthcare services Because the VSS and healthcare facilities pursue distinct objectives, the latter may not always behave in the VSS's best interests.
From a principal-agent perspective, reimbursement methods may impact the number of healthcare services provided by healthcare facilities to insured individuals To be specific, under the fee-for-service payment method, healthcare facilities are reimbursed retrospectively for each service they provide (Langenbrunner et al.,
2009) As this method relates healthcare facilities' income to the number of services delivered, it creates an incentive for them to increase the number of consultations or offer more services to each patient (Guinness & Wiseman, 2011) If this is the case, the effect of health insurance on the use of healthcare may be overestimated By contrast, with the capitation approach, providers are paid in advance at a fixed rate to offer a certain set of services to an individual for a specific period (often one month or one year) Because of this, hospitals have the incentive to lower the total number of healthcare treatments per insured person (Langenbrunner et al., 2009) In this situation, the insured may decrease their hospital visits as a result of obtaining fewer treatments, consequently mitigating the impact of health insurance on healthcare utilization.
Theory of precautionary savings
The theory of precautionary savings was developed by a multitude of scholars, such as Leland (1968), Drèze & Modigliani (1972), Sandmo (1970), Hubbard et al.
(1995), and Kimball (1990b) The basic implication of the model is that individuals, in the presence of uncertain future income, are likely to diminish consumption and save more Leland (1968) suggested for the first time that precautionary savings are linked to uninsurable (background) risk Kimball later introduced the concept of
“prudence” as a measure of the intensity of precautionary saving motives (M S. Kimball, 1990) The effect of background risk on savings and consumption, from his perspective, is determined by the sign of the third derivative of the Neumann- Morgenstern utility function In other words, the marginal utility functions of prudent persons are convex and have positive third derivatives This kind of person tends to achieve greater current savings and reduce current consumption when facing increasing uncertainty about future resources This is attributed to the fact that with positive third derivatives, prudent people’s expected marginal utility of saving climbs as they face a rise in the background risk (Noussair et al., 2014). Precautionary savings can be thought of as the amount of wealth people use to protect themselves against background risks in the future Importantly, Kimball also suggested the notion of decreasing absolute prudence (M Kimball, 1990) For an individual with decreasing absolute prudence, his precautionary motive of saving is a decreasing function of his wealth, meaning that if he becomes wealthier, he is less sensitive to the risk.
Health expenditures can be viewed as a considerable source of future uncertainty for a household since they might be substantial relative to income, persistent, and positively correlated with age Based on the logic of the theory of precautionary savings, participation in the health insurance scheme through reducing unexpected out-of-pocket health expenses may significantly diminish the need for precautionary savings and accordingly raise current consumption (Chou et al., 2003) It is, however, crucial to distinguish between the two conflicting effects of health insurance on savings and consumption On the one hand, its risk effect raises savings but reduces current expenditures On the other hand, in terms of the income effect, it could be argued that because health insurance is viewed as an income transfer, it provides opportunities to increase both consumption and savings(Chou et al., 2003; Kirdruang & Glewwe, 2018) However, savings do not exist in a vacuum; rather, they join a variety of different instruments, including private insurance, investments, and credit, that households use in their financial management strategies Using this logic, enrollment in a health insurance program impacts not just households’ savings but also their usage of other financial services.
Figure 2.3 depicts the conceptual framework of the thesis Grossman’s model suggests that the demand for healthcare is derived from the demand for health It is expected that after the implementation of Vietnam's free health insurance program for the elderly, the relative decrease in out-of-pocket prices for healthcare services will result in an increase in healthcare utilization Besides, the bottom of the framework depicts how Andersen's behavioral model explains the effect of health insurance coverage on healthcare usage Under the model, health insurance is viewed as an enabling factor that, in combination with predisposing and need factors, boosts healthcare utilization From the perspective of moral hazard theory, having health insurance generates an ex-post moral hazard effect, which drives up healthcare utilization.
It may be deduced from the economic theory of moral hazard and cost-sharing that a co-payment policy will reduce both desirable and undesirable moral hazards. Reduced desirable moral hazard is linked to reduced essential care, which has a negative impact on health.
The theory of precautionary savings suggests that once enrolled, health insurance will provide financial protection for a household, hence affecting the choice of savings and other financial tools.
Figure 2.3: Conceptual framework for the thesis Source: Own construction
Health Economic theory of moral hazard and cost-sharing
Program impact measures
There are a range of program effect measures, and the choice of which measure to employ will depend on the policy topic of interest A few of the measures are described below.
ATE is the average effect of program participation for the entire population, and it is calculated as follows:
ATE = E[Y 1i − Y Oi ] where Y 1i and Y Oi represent the outcomes for the treated and control individuals.
This estimate assumes that all people allocated to a program are compliers Thus, ATE is often applicable to mandated programs in which everyone participates. However, even though the programs are mandatory, it is common for not everyone to participate; there exist some who participate but drop out before the program is completed Furthermore, in voluntary programs, it is not feasible to force everyone to participate In these circumstances, those who enroll in the program vary from those who do not in a variety of ways, and these differences probably also have an impact on the probability that they will participate The mean difference in outcome between the treated and control groups no longer represents the ATE but rather the so-called intention-to-treat effect (ITT) The ITT compares outcomes between those who received treatment and those who did not, regardless of actual enrollment (Alkenbrack, 2011).
Average Treatment Effect of the Treated (ATT)
Because programs often reach just a subset of the population, policymakers are typically interested in the average effect of program participation for existing program participants (ATT).
ATT = E(|Y 1i − Y Oi |T = 1) where T ∈ {0,1} is a binary indicator of treatment status The ATT will reflect the impact of the treatment on those who got it, independent of treatment or control group membership (Garrido et al., 2016) In a well-designed randomized controlled trial with complete compliance to treatment assignment, the ATE and ATT should have the same value and indicate the treatment's effectiveness since randomization should guarantee that the control group is comparable to the treatment group (Garrido et al., 2016).
Local Average Treatment Effect (LATE)
When implementing a program, policymakers may concentrate on a segment of the population that was previously non-participant but changed their status owing to the program's execution These individuals are called compliers The LATE estimates the average impact on compliers.
E(∆|S € ⊂ N) = E(Y 1i − Y Oi |S € ⊂ N) where S € signifies the group of compliers who convert from nonparticipation to participation as a result of the program's execution N is the nonparticipant group.
Program impact evaluation methods
The measurement of the causal effects of health programs is one of the research priorities in health economics Researchers are interested in determining the effect a person's participation in a program has on a certain outcome variable Program effect evaluations were primarily guided by the works of Holland (1986) and Rubin
D B (1974) Let Y 1 represent the outcome if an individual participates in the program
(receives treatment), and Y O represent the outcome if he does not The causal effect of the treatment on individual i is equal to the difference between the two outcomes, such that:
The problem in program evaluation is that, in practice, each individual is in either the treated group or the untreated group As a result, only one of the two outcomes is observed, and the causal effect cannot be estimated If an individual participates in the program, then Y 1 is observed (factual), but Y O is not (counterfactual). Similarly, if an individual is not exposed to the treatment, only Y O (factual) can be seen; in this instance, Y 1 is the outcome of him if he had been treated (counterfactual) Holland (1986) referred to the inability to observe an individual in both of his two states as the "fundamental problem of causal inference."
Given the fundamental problem, one strategy for determining the counterfactual outcome is to use the outcome of the non-participants as a substitute for the outcome that the participants would have had in the absence of treatment Only if participants and non-participants had identical characteristics would this strategy be valid However, the program is not assigned randomly but rather based on the needs of communities and individuals who then self-select the program (Khandker et al.,
2009) As a result, participants have different characteristics that make them more likely to join the program than non-participants and also impact their outcomes. Selection bias is the term used to describe this phenomenon (Becker & Caliendo,
2007) Thus, endogenous program uptake should thus be taken into account when estimating the causal impact of an insurance program on the outcomes of interest.The use of experimental and quasi-experimental approaches is motivated by the goal of eliminating selection bias.
RCTS involve randomly assigning individuals to a group that receives the treatment (the treated group) and a group that does not (the control group) Random assignment enables the formation of a control group with the same distribution of observed and unobserved characteristics as the treated group (A Smith & E Todd,
2005) If RCTS are conducted properly, differences in outcomes between the treated and control groups may be attributed to the treatment, and no confounding variables impacting the outcomes of interest can have influenced treatment assignment, hence removing selection bias (Bọrnighausen et al., 2017) RCTS are regarded as the “gold standard” for assessing the causal effects of programs and policies (Kim & Steiner,
2016) RCTS, however, are not always practicable owing to ethical, operational, and political concerns In this situation, researchers can rely on the next best approach, called quasi-experiments.
A quasi-experimental approach is different from a randomized controlled trial in that individuals are not randomly assigned to a treated or control group Due to the susceptibility to bias and confounding introduced by the lack of random assignment inherent in quasi-experimental approaches, the scientific credibility of results from these designs is lower than that of RCTS Nevertheless, well-conducted quasi- experimental research may yield compelling evidence for causal inference (Rockers et al., 2015).
A quasi-experimental design identifies a comparison group with baseline characteristics as comparable to the treatment group as possible (White & Sabarwal,
2014) The comparison group reflects what the outcomes would have been if the program or policy had not been implemented; hence, the program or policy can be considered to have caused any difference in outcomes between the treatment and comparison groups (White & Sabarwal, 2014) A range of widely used quasi- experimental techniques provide reliable causal inference when properly implemented, as follows:
DID is among the most widely utilized approaches in studies of impact evaluation.
By comparing the outcomes of those who were treated with those who were not treated before and after the policy was implemented, the DID method controls for bias from both the observed and unobserved time-invariant variables Thus, this method needs data from both before and after the policy in the form of panel data or repeated cross-sectional data Let Y 1it denote the outcome if individual i participates in the program at time t, and Y Oit denote the outcome if he does not at time t, the DID estimate of a policy's effect is as follows:
The DID method's most significant weakness is its reliance on the parallel trends assumption The assumption implies that in the absence of policy, both the treated and the control groups will exhibit identical trends in the outcome variable Because the treated group is only seen in its treated state, the assumption cannot be tested (Fredriksson & Oliveira, 2019).
The PSM involves constructing and matching treatment and comparison groups with similar observed characteristics When employing PSM, researchers have to follow steps that include choosing the suitable set of confounders, estimating the propensity score, and then utilizing matching techniques like nearest neighbor matching, radius matching, or kernel matching to match the treated group with the control group.
Conditional independence and common support are two assumptions that must be satisfied in order to apply the PSM estimator The conditional independence assumption states that after observed characteristics have been accounted for, the outcome is no longer related to the treatment status other than via the program's impact (Binci et al., 2018) The primary disadvantage of PSM is that it depends on matching people based on their observed characteristics Therefore, if there exist unobserved factors that influence participation, the estimates will be biased The assumption of common support implies that the estimated propensity score for all individuals in the treated and control groups must be between 0 and 1 (Binci et al., 2018).
In the econometrics literature, instrumental variable models are the standard method for isolating the impact of treatment in the presence of unobserved confounders (Chib & Greenberg, 2007) The basic steps to implement the IV method are first to identify a variable (instrumental variable) that affects treatment assignment but has no direct effect on the outcome; second, use this variable to find variation in the treatment that is independent of unobserved confounders; and third, use this variation in the treatment that is independent of unobserved confounders to evaluate the treatment's causal effect (Baiocchi et al., 2014) The variable Z is an instrument if it satisfies the following three assumptions: relevance, exclusion restriction, and exchangeability The relevance assumption requires that instrument Z must have an impact on exposure X; the exclusion restriction assumption implies that any effect of the instrument Z on the outcome Y is exclusive via its effect on exposure X; Exchangeability indicates that the impact of Z on Y is not confounded (Lousdal, 2018).
The most challenging part of the IV technique is finding an IV that meets all three of the abovementioned assumptions (Brookhart et al., 2010) An IV is considered to be weak if it is not a strong predictor of exposure (Brookhart et al., 2010) If the correlation between the IV and the exposure variables is weak, then the standard error and bias will both be higher Another limitation of the IV technique is that it is impossible to directly test the exclusion restriction assumption with the data at hand.
Review of key findings
1.3.2.3 Impacts of health insurance on healthcare utilization and out-of-pocket expenditures
There has been a substantial body of research on the relationships between health insurance and healthcare utilization in various populations and settings The relationship directions are ambiguous Many studies centering on the impact of health insurance on healthcare utilization observe a positive effect Others, however, demonstrate no, limited, or even negative effects Freeman et al (2008) conducted a systematic review of empirical evidence of causal links between health insurance and healthcare consumption among non-elderly individuals in the United States (US), based on research published after 1991 The review includes 14 studies, and the findings indicate that health insurance is associated with greater use of medical services and preventative care Likewise, Spaan et al (2012) conducted a systematic review of research completed in Africa and Asia One hundred fifty-nine research articles published between 1990 and 2011 match the criteria, and the evidence suggests that health insurance increases care use A recent review of 26 papers from all over the world, however, shows that private insurance did not cause more people to use healthcare (C Zhang et al., 2020) Furthermore, another systematic review of factors influencing healthcare usage in China indicated that health insurance coverage had a negative impact on healthcare utilization in 6 of 16 studies (S. Zhang et al., 2019).
Regarding the effect on out-of-pocket expenditures, previous studies have also shown varied findings Erlangga, Suhrcke, et al (2019) conducted a systematic review and found that 34 of the 46 studies examined the effects of health insurance on out-of- pocket health expenditure 17 of the 34 studies showed a decrease in out-of-pocket expenditure, 15 found no statistically significant impact, and two studies—one each from Indonesia and Peru—indicated a rise in out-of-pocket expenditure.
Even though there are many studies on the impacts of health insurance on healthcare utilization and out-of-pocket expenditure, there are few that focus on the elderly in LMICs Most of the results from the limited research available indicate that health insurance enhances the utilization of healthcare (Cheng et al., 2015; Fink et al., 2013;
H Liu & Zhao, 2012; Mao et al., 2020; Qiu & Wu, 2019; Tungu et al., 2020; Z. Wang et al., 2018; Zhou et al., 2020) With reference to the impact on out-of-pocket expenditure, findings from studies centered on the elderly are similar to those in other population subgroups Limited or no evidence of the impact was identified in some studies (Cheng et al., 2015; Fink et al., 2013; H Liu & Zhao, 2012; Z Wang et al., 2018) Others, on the other hand, reported a reduction in out-of-pocket expenditures for insured persons compared to those without insurance (Aryeetey et al., 2016; Ku et al., 2019; A Zhang et al., 2017) In particular, Wang et al (2018) investigated the effects of social health insurance on healthcare utilization and costs in Chinese middle-aged and elderly community-dwelling adults The findings reveal that those with social health insurance spend more on total and out-of-pocket healthcare than people without social health insurance, showing that participation in social health insurance causes a considerable rise in total and out-of-pocket healthcare expenditure.
In summary, the current understanding of the impacts of health insurance on healthcare utilization and out-of-pocket expenditure of the elderly in LMICs is inadequate at the moment due to a scarcity of literature and inconsistent findings from current studies Mixed and limited findings rationalize the need to further investigate the impacts of health insurance on the elderly In this regard, the present thesis seeks to cover some such gaps by evaluating the effects of health insurance on healthcare utilization and out-of-pocket expenditures of the elderly in Vietnam.
When examining the effects of insurance on the elderly, a majority of prior research in LMICs included everyone aged 60 or older or 65 or older into a single treatment group (Cheng et al., 2015; Erlangga, Suhrcke, et al., 2019; Fink et al., 2013; H Liu
& Zhao, 2012; Z Wang et al., 2018) The response to insurance of a treatment group that includes everyone over the age of 60 or 65 may not adequately reflect the response of a certain sub-elderly population It is worth mentioning that Mao et al.
(2020) utilized data from China's National Health Service Surveys to examine how health insurance affects the use of health services by older adults in urban China. They discovered that individuals aged 70–79 were more likely than those aged 60–
69 to have outpatient visits and hospital admissions, whereas those aged 80 and above had no significant difference Also, in China, Qiu & Wu (2019) investigated the effects of three basic medical insurance schemes on outpatient visits by the elderly They also separated the elderly into three groups based on their ages: 60 to
69, 70 to 79, and 80 and older The findings showed that insurance plans had varying impacts on three distinct age groups.
The different effects of health insurance on different older age groups in China can be evidence that each age group reacts differently to insurance To get a full range of evidence, future studies should focus on the subgroups of the population for which there is still little or no evidence This thesis is intended to address the knowledge gap by examining the impact of insurance on the healthcare utilization and out-of-pocket expenditures of the middle-old (75–84 years old) as defined by Zizza et al (2009) This study is particularly important in the Vietnamese context, where the elderly have a legal right to priority medical examination and treatment (The National Assembly of Vietnam, 2009) Nevertheless, no study has directly examined the impact of health insurance on the elderly group.
1.3.2.4 Impacts of health insurance on households’ financial service choices
Although it is not much, there has been some research done around the world to see how health insurance affects each financial service, such as savings, credit, and life insurance Most of the studies were done in The US context Bornstein & Indarte,
(2020) exploited the staggered expansions of one of the major US social insurance programs, Medicaid, to estimate the causal impact of expanding insurance on household debt According to the findings, increasing Medicaid led to a 2.2% rise in borrowing Also, in The US, Gallagher et al (2020) evaluated the impact of Medicaid access on household savings behavior They discovered significant heterogeneity in the savings response to Medicaid among households Households who were not suffering financial difficulty acted in agreement with a precautionary savings model, which indicates they saved less under Medicaid In contrast, among low-income households, Medicaid participation improved return savings rates by around 5 percentage points, or $102 Kumar (2019) found that life insurance choice was strongly connected with health insurance choice, and these two options were mutually reinforcing in families' risk-minimizing goods baskets in the U.S.
Using a synthetic panel data approach, Kirdruang & Glewwe (2018) investigated the impact of Thailand's Universal Health Coverage Scheme on household consumption and savings Difference-in-differences estimates showed that the Universal Health Coverage Scheme had little or no effect on either savings or consumption expenditures in the short run Savings were unaffected by the introduction of universal coverage in the long run, although consumption did rise.
Because health insurance helps individuals to reduce unexpected out-of-pocket costs, it is considered among the financial options that households choose for their risk management, and thus, it correlates with other services such as private insurance, savings, investments, and credit A conventional way to investigate the impact of health insurance on financial services that households need for their risk management strategies is to estimate them in separate single equations In developing countries, households generally consider including a group of financial services such as private insurance, savings, investments, and credit in their financial risk management strategies Hence, it would be a reasonable expectation that there exist unobserved factors that simultaneously affect financial services and thus establish correlations among savings, insurance, and loans And because of that, using single-equation models like the above-mentioned studies that do not account for the inter-linkages may produce inefficient and inconsistent coefficients.
The review of the literature suggests that there have been a scant number of authors conducting research into the impact of health insurance on savings and other household financial services Additionally, in developing countries where households often include several financial services in their risk management strategies; thus, the correlations between these services should be taken into account However, until the present time, no study across the world has been conducted to simultaneously examine the impacts of health insurance on services such as private insurance, savings, investments, and credit Thus, it emphasizes the importance of more research into the impact of health insurance on concurrent financial services in order to present a more detailed picture of the effects and to account for their correlations.
1.3.2.5 The impact of cost-sharing on health
IMPACTS OF HEALTH INSURANCE ON HEALTHCARE
Model specification
This section demonstrates the ways that the RDD is employed to verify the causal effects of the health insurance program for the elderly on healthcare use in rural Vietnam The primary interest in this research is the relationship between healthcare utilization and insurance status The simple model that can represent this relationship is as follows:
Here, the dependent variable Y i denotes the outcome of interest, being the number of hospital visits or the probability of a hospital visit or expenditures per visit for individual i; D i is a binary variable for individual i to be insured by health insurance; X ' is a vector of observed characteristics of individuals that might affect the outcomes; u i is a disturbance term The coefficient of interest in this equation is y 1 , which measures the effect of insurance coverage on healthcare consumption, after
Health insurance Moral hazard effect
Grossman’s model Demand for healthcare i controlling for observed characteristics of individuals However, the possible endogeneity of the health insurance status makes it less likely to produce a consistent estimate of y 1 Especially, there still exist potential unobserved confounders affecting the relationship between health insurance status and outcomes.
In this research, the RDD is adopted to solve the problem of endogeneity of the health insurance status and obtain an unbiased estimate of y 1 The RDD was first developed by Thistlethwaite & Campbell (1960) as a method to investigate causal effects in cases where a random assignment was not possible This approach has been widely used to investigate program effects in an extensive diversity of economic contexts (Lee & Lemieux, 2010) This study’s RDD exploits the exogenous discrete jump in the probability of insurance take-up arising from the health insurance program for the elderly Given the fact that old people have no manipulation of their age, the health insurance program for the elderly generates an exogenous variation in insurance coverage in the vicinity of 80 years of age Since the probability of participation in the program does not switch from 0 to 1 around the cutoff value of age (imperfect compliance), the fuzzy RDD is applied (Lee & Lemieux, 2010) In this design, the ratio of the jump in the outcome to the jump in the probability of treatment at the cutoff from both sides is interpreted as an average causal effect of the treatment (Imbens & Lemieux, 2008) Formally, the estimand is
FRD Effect of threshold on
As indicated by Lee & Lemieux (2010), the regression model for the fuzzy RDD basically includes two stages as follows:
Where D i is a binary variable of insurance status for individual i D^ i is the predicted value of the probability of being insured obtained from equation (3.2) T i is a binary variable that indicates whether an individual belongs to the treatment group (age ≥ 80 years), and serves as an instrumental variable (IV) for insurance status age i refers to the age of an individual f(.) is the smooth function of age and health insurance status; g(.) is the smooth function of age and the outcomes X ' is a vector of control variables. v i and e i are random error terms.
Two-stage least squares (2SLS) estimation is applied to generate fuzzy RDD estimates To be specific, the first-stage equation is estimated using a linear probability model Coefficient α 1 identifies the shift in the probability of being insured for an elderly person due to exposure to the policy For the second-stage, predicted values of the probability of being insured (D^ i ) in the first-stage will be used for regressions in the second-stage equation Because the demand for the healthcare of the elderly differs with age, the application of linear models will lead to bias in the treatment effect Therefore, to minimize bias attributed to the misspecification of the functional form, this study follows Palmer et al (2015) to estimate the second-stage equation with some nonlinear models depending on the type of outcome variable Specifically, the Probit regressions are applied to the probability of a hospital visit, and the Poisson and Tobit models are respectively applied to the number of visits and expenditures per visit.
In equation (3.3), the important coefficient of interest is β 1 , which captures the local average treatment effect (LATE) (Imbens & Angrist, 1994) of the health insurance program for the elderly on healthcare utilization LATE is the average treatment effect of compliers whose health insurance status would switch from uninsured to insured if their ages satisfy the cutoff rule (age i ≥80), but would not otherwise be insured (Lee & Lemieux, 2010) In addition, the LATE is only valid within a population whose running variable values lie within a region close enough to the threshold for individuals to be considered exchangeable (O’Keeffe & Baio, 2016).
Therefore, in this research, the impact of the policy is just valid to individuals whose ages are in close proximity to 80 years old, which mainly belong to the middle old.
Both parametric and non-parametric strategies are employed for the estimation of the fuzzy RDD In the baseline model, a parametric strategy is elected with a ten- year bandwidth on either side of the cutoff point The parametric approach has its own pros and cons It uses all data from both close to and far from the discontinuity and thus contributes to the precision of the treatment effect However, the results may be driven by noisy data points that are far away from the threshold of 80 years and be sensitive to outliers f(.) and g(.) will be applied under linear forms This study does not apply higher order of running variable (age) as Gelman & Imbens
(2019) noted that estimates based on high-order polynomial regressions are sensitive to the order of the polynomial, and the inference based on these settings is often poor.
To relax the possibility of bias associated with using a large range of data (Jacob et al., 2012), the non-parametric strategy is performed in the robustness check By utilizing a much smaller portion of the data around the cutoff point, the non- parametric method minimizes the chances of bias but decreases the statistical power due to the smaller sample size (Jacob et al., 2012) While in the baseline model, a wide bandwidth of 70-89 years old is used, in the robustness check, some smaller different bandwidths are picked The bandwidth is reduced starting from 72-87 years old to 74-85 years old and subsequently to 75-84 years old, which is half the bandwidth in the parametric technique In addition, ignoring non-compliance, this research also estimates the coefficient y 1 in the reduced form equation (3.1), which measures the intention-to-treat (ITT) effect The estimated results will serve as another robustness check.
Data and variables
The data for this study is drawn from the Vietnam Household Living StandardSurvey (VHLSS) The survey was biennially conducted in 63 provinces and cities nationwide and was representative of the entire population of Vietnam In the baseline analysis, this study uses four waves of VHLSS data in the years 2012, 2014, 2016, and 2018, which were all fielded after the health insurance program effect for the elderly The data is cleaned by deleting duplicates and dropping observations with null information In the baseline model, the sample is limited to the elderly aged 70 to 89 years old living in rural areas, yielding a number of observations of 1,011, 1,043, 1,079, and 5,453 for four years, respectively (sample size: 8,586) In the robustness check, the sample sizes for bandwidths of 72-87, 74-85, and 75-84 are 6,845, 5,204, and 4,372, respectively Additionally, the 2010 wave, a pre-policy dataset, is used for a placebo check All monetary values are in thousands of Vietnamese Dong and adjusted to the prices of 2012, using the consumer price index.
In fuzzy RDD, control variables can be used to reduce bias and improve the precision of estimates (Imbens & Lemieux, 2008) The controls in this study are pre-determined characteristics of individuals comprising age, retirement status, gender, marital status, ethnicity, income, education, region, and year-fixed effects (see Table 3.1 for a full and detailed list of control variables used in this research). The outcome variables of interest in this study comprise six indicators of healthcare utilization, which include the probability of an outpatient visit, the probability of an inpatient visit, the number of outpatient visits, the number of inpatient visits, expenditures per outpatient visit, and expenditures per inpatient visit (see Table 3.2 for definitions of outcome variables).
Table 3.1: Descriptive statistics of control variables
Definitions of variable 70≤Age≤79 80≤Age≤89 Differences
Mean (S.E) Mean (S.E) Mean (S.E) D: Health insurance status (1=if insured, 0= if uninsured) 0.886 (0.004) 0.961 (0.004) 0.0747*** (0.006)
Retiree (=1 if having monthly pension, = 0 if no) 0.145 (0.005) 0.101 (0.005) -0.0440*** (0.008)Male (1=if yes, 0= if no) 0.449 (0.007) 0.412 (0.009) -0.0371*** (0.011)
Married (1=if currently married, 0= if others)
Kinh majority (if belonging to Kinh ethnic group,
Income quartiles (of household average income)
1st income quartile (the lowest) (1=if yes, 0= if others) 0.246 (0.006) 0.258 (0.008) 0.0126 (0.010)
2nd income quartile (1=if yes, 0= if others) 0.242 (0.006) 0.264 (0.008) 0.0214** (0.010) 3rd income quartile (1=if yes, 0= if others) 0.258 (0.006) 0.236 (0.008) -0.0214** (0.010) 4th income quartile (the highest) (1=if yes, 0= if others) 0.254 (0.006) 0.242 (0.008) -0.0126 (0.010)
No degree (1=if yes, 0= if others) 0.448 (0.007) 0.629 (0.009) 0.1813*** (0.011) Primary school (1=if yes, 0= if others) 0.279 (0.006) 0.240 (0.008) -0.0393*** (0.010) Lower secondary school (1=if yes, 0= if others) 0.147 (0.005) 0.064 (0.004) -0.0828*** (0.007) Upper secondary school (1=if yes, 0= if others) 0.030 (0.002) 0.016 (0.002) -0.0140*** (0.004) Vocational school (1=if yes, 0= if others) 0.072 (0.003) 0.038 (0.003) -0.0345*** (0.005) College or above (1=if yes, 0= if others) 0.023 (0.002) 0.013 (0.002) -0.0107*** (0.003)
Red River Delta (1=if yes, 0= if others) 0.284 (0.006) 0.328 (0.008) 0.0443*** (0.010) Northern midlands and mountainous areas (1=if yes, 0= if others) 0.124 (0.004) 0.104 (0.006) -0.0204*** (0.007) North Central and Central coastal areas (1=if yes, 0= if others) 0.248 (0.006) 0.280 (0.008) 0.0326*** (0.010)
Central Highlands (1=if yes, 0= if others) 0.033 (0.002) 0.033 (0.003) -0.0003 (0.004) South East (1=if yes, 0= if others) 0.073 (0.003) 0.059 (0.004) -0.0140** (0.006) Mekong River Delta (1=if yes, 0= if others) 0.239 (0.006) 0.197 (0.007) -0.0421*** (0.009)
Notes: The data comes from VHLSS 2012, 2014, 2016, and 2018 Significance levels: * < 10% ** < 5% ***
Table 3.1 outlines the descriptive statistics of independent variables for people aged
70 to 79 years and 80 to 89 years Nearly 89% of individuals aged under 80 are insured, compared to roughly 96% of individuals aged 80 or above A gap is seen in marital status between the two groups; approximately 63% of individuals younger than 80 years are currently married, compared to 42% of individuals aged 80 years or above In terms of Retiree, Male, and Kinh majority, there exist slight differences between the two groups With regard to Income quartiles, the rates of individuals are equally distributed between two groups and across four quartiles of income. Specifically, each quartile in both groups contains around 25% of the people in the sample As for Education level, 45% of the people in the control group have no degree while the percentage in the treatment group is 63% Excepting Nodegree, at each education level, the rate of the treatment group is basically lower than that of the control group.
Table 3.2 depicts the means of outcome variables for the elderly in the two groups along with the differences It is worth noting that, in general, no substantial differences are observed between the two groups Individuals aged 80 years old or above have a higher probability of an inpatient visit and record more frequent inpatient visits than those under 80 People aged 70-79 pay a lower cost for an outpatient visit than those aged 80-89 And surprisingly, the elderly in the treatment group are less likely to have an outpatient visit and have a smaller number of outpatient visits than the people in the control group.
Table 3.2: Descriptive statistics of outcome variables
Definitions of variable 70≤Age≤79 80≤Age≤89 Differences
Mean (S.E) Mean (S.E) Mean (S.E) Probability of an outpatient visit: Probability of an outpatient visit to medical establishments over the past 12 months (1=if yes, 0= if no) 0.59 (0.01) 0.56 (0.01)
Probability of an inpatient visit: Probability of an inpatient visit to medical establishments over the
- 0.039*** (0.011) past 12 months (1=if yes, 0= if no) 0.21 (0.01) 0.24 (0.01) 0.024*** (0.009)
Number of outpatient visits: Number of outpatient visits to medical establishments over the past 12 - months
Number of inpatient visits: Number of inpatient
3.20 (0.10) 2.74 (0.09) 0.463*** (0.147) visits to medical establishments over the past 12 months 0.38 (0.01) 0.45 (0.02) 0.063*** (0.024)
Expenditures per outpatient visit: the average costs for an outpatient visit 218.30 (10.64) 187.18 (11.95) -31.113* (16.847) Expenditures per inpatient visit: the average costs for an inpatient visit 757.93 (56.13) 794.81 (74.96) 36.882 (93.870)
Notes: The data is drawn from VHLSS 2012, 2014, 2016, and 2018 Significance levels: *