Health Insurance And Health Care Access In China

Because private providers do not accept insurance, the insurance scheme may cause patients to shift usage from private to public providers, thus not necessarily resulting in an overall e

HEALTH INSURANCE AND HEALTH CARE ACCESS IN CHINA

A Thesis submitted to the Graduate School of Arts & Sciences

at Georgetown University

in partial fulfillment of the requirements for the

degree of Master of Public Policy

in the Georgetown Public Policy Institute

By

Verinda Jean Esther Fike, B.A

Washington, DC April 14, 2008

HEALTH INSURANCE AND HEALTH CARE ACCESS IN CHINA

Verinda Jean Esther Fike, B.A

Thesis Advisor: Michael Clemens, Ph.D

ABSTRACT

The Chinese government has made recent efforts to expand health insurance to rural areas that have been primarily dependent upon private health care providers Because private providers do not accept insurance, the insurance scheme may cause patients to shift usage from private to public providers, thus not necessarily resulting in

an overall expansion of health care delivery This study uses a 2001 survey of 3,600 households in three Chinese provinces to analyze whether health insurance is actually expanding overall service delivery or is simply switching patients from private to public provider utilization By statistically comparing the provider choices of

households with health insurance to those of uninsured households, this study finds evidence that health insurance expands overall health care utilization in China The

findings relate only to service utilization and do not address health outcomes

TABLE OF CONTENTS

INTRODUCTION 1

LITERATURE REVIEW 4

Historical Background of Health Care Policy in China 4

Rural vs Urban Health Care 6

Health Insurance 8

The Quality of Private Health Care in China 12

CONCEPTUAL MODEL 13

DATA DESCRIPTION 16

ANALYSIS PLAN 19

Dependent Variables 19

Independent Variables 22

Interaction Terms 26

Discussion 28

DESCRIPTIVE STATISTICS 31

RESULTS 33

DISCUSSION 42

Interactions 46

Patient preferences: Demand vs Supply 51

CONCLUSION 56

REFERENCES 58

INTRODUCTION

China is facing a health care crisis While total health care spending in China is increasing, the government’s share of this total spending has shrunk by more than half since 1980 This decrease in public health financing has been devastating for rural households where more than 90% of health care spending is now out-of-pocket Although 70% of the population lives in rural areas, public health expenditures in these areas constitutes only 30% of the national total (Lindelow and Wagstaff, 2003) In some parts of China, more than 60% of those in dire poverty have been driven there by these huge out-of-pocket medical expenses (Liu et al., 2003)

The disparity between basic health conditions in urban and rural areas of China

is enormous In 1999, infant mortality was 37 per 1,000 live births in rural areas, as compared with 11 per 1,000 in urban areas (Blumenthal and Hsaio, 2006) Perhaps most shocking is that in some poor rural areas, infant mortality has recently increased While broad outcomes of this kind depend on much more than health care provision, these numbers nevertheless suggest that access to quality medical care for rural

residents remains limited and deficient

In an effort to make health care more available at lower cost, the Chinese government in March of 2007 announced plans to expand its new cooperative health care system to cover all rural counties by 2010 (Xinhua, 2007) The expansion of the program is expected to cost US$1.3 billion this year and US$750 million for each

subsequent year This program will allow rural residents to use a joint fund to pay for visits in public facilities, but the fund will cover only 40-60% of medical bills The joint fund will not be accepted at private health care facilities

New Chinese health care policies generally fail to recognize the vast number of private health care facilities serving both rural and urban areas Private health care facilities are often more accessible and less costly than public facilities (Liu et al., 2006) Even if a particular treatment is more affordable at a public facility, the cost and hassle of transportation to a public provider that is farther away than a private facility may outweigh the advantage of having insurance However, urban and rural residents may experience this difficulty differently as urban residents have greater access to public facilities

Previous research suggests that the quality of care in private facilities is better, worse, or the same as public facilities (Meng et al., 2000; Lim et al., 2002) The rising incomes of rural households may therefore influence households with health insurance

to seek care at public or private facilities, depending on whether they believe one to have superior care over the other For more serious illnesses, however, households may choose public providers which are generally better equipped to handle these illnesses

Although some policymakers believe that expanding health insurance coverage for rural households is considered effective in making health care more affordable and thereby expanding overall coverage, the results of any such effort depend crucially on patients’ decisions to utilize facilities where they are covered (public) versus those

where they are not (private) This paper seeks to understand whether health insurance

is expanding overall coverage or whether it is simply moving people from private health care provision to public provision Using data from a 2001 household survey of 3,600 households in three provinces in China, this study sets out to answer these questions This study finds that households with health insurance in large measure choose public providers This outcome is greater than the deterrent outcome of not choosing a private provider, suggesting that health insurance is increasing overall service delivery If redistribution is the goal of the Chinese health insurance program, then it appears this goal is being achieved through the scheme A policy to extend the program to other parts of China should therefore be considered

If the goal of the health insurance program is to improve the overall health of the population, then more research is needed Expansion of service provision does not necessarily mean that health outcomes are improving, among other things because those being brought into the system could be those with the worst health and hopeless conditions One way to learn more about the relationship between health insurance and health outcomes would be to make insurance mandatory within a delimited group

on a pilot basis and carefully track their health status over a number of years relative to

a similar comparison group

In order for either of these policy goals to be met, households should have access to health care facilities where their insurance may be used Therefore, the final recommendation of this study is to increase access to households by building more public facilities in rural areas and/or by allowing insurance to be accepted at private

facilities Because rural areas are predominantly served by private providers, the latter option is likely to be the least costly

LITERATURE REVIEW Historical Background of Health Care Policy in China

China is a developing country struggling to meet the global demands of

privatization while continuing to claim socialist status The many economic reforms the country has implemented have also created volatile social cleavages and challenges within China as urban coastal areas have benefited from the massive growth of the economy more rapidly than inland rural areas This division is particularly evident in the health care system where rural areas continue to be plagued chronic issues Not only is infant mortality over three times higher than in urban areas, but HIV/AIDS is spreading more rapidly among rural people Tens of thousands of peasants are

estimated to have contracted the disease during blood collection at unsanitary rural clinics (Chan, 2001) While the health care system in urban areas is far from perfect, the system’s problems are intensified in the countryside

China has undergone repeated health care reforms since the 1980s The

massive changes within the health care system have shaped and been shaped by the extraordinarily rapid economic growth of the country China’s “growth-first” strategy during the 1980s under Deng Xiaoping emphasized economic growth at any cost (Meng et al., 2004) The health care reforms under this strategy included dramatically cutting public subsidies to hospitals and implementing new rules which allowed

providers to charge patients more than average cost for certain services, such as

prescription drugs and high technology diagnostic procedures—a similar tactic

executed in the early years of U.S health care reform (Wang, 2004)

This pricing scheme had two main consequences First, public hospitals

experienced distorted incentives, viewing high technology and prescription drugs as their only means to meet the demands of decreasing government subsidies and

increasing budgets (Eggleston, 2006) This allowed physicians to over-stress the importance of specific services, often giving unnecessary treatment and over-

prescribing drugs

Second, health care reform gave rise to the re-emergence of private hospitals and clinics, which began to flourish under this new system China has a history of

“barefoot doctors” who provided services outside the public health realm in the

countryside, but these doctors became virtually nonexistent during the Cultural

Revolution of the 1960s (Blumenthal and Hsiao, 2005) The 1980 economic reforms, however, encouraged privatization, and many private facilities resurfaced in rural areas where barefoot doctors once provided services As public hospitals responded to budget cuts by seeing more patients, private facilities were often able to out-compete public ones, due to their lower infrastructure and staffing costs Despite the fact that these private doctors in China often have very little formal medical training, the

government allowed these private facilities to exist, given the demand for health care services in rural areas (Wang, 2006) Although Chinese health care guidelines apply to both public and private service providers, government oversight of private health care

facilities has been criticized as being too lax Because private providers are

concentrated in rural areas, this lack of government oversight is particularly

problematic in the countryside

Rural vs Urban Health Care

Health care reform has generated different effects in urban and rural areas and

is partially to blame for the divergence in health outcomes in these areas As in many other developing countries, both urban and rural health care providers in China are primarily paid for by fee-for-service (FFS) which has been associated in OECD

countries with producing higher health expenditures as a fraction of total GDP

(Eggleston et al., 2006) In China, the FFS payment system when combined with a distorted fee schedule is widely acknowledged to spur cost escalation As income in rural areas is much lower than urban areas, the FFS system is shown to have a negative effect upon rural areas in many developing countries (Eggleston et al., 2006)

While a vast number of private health care facilities serve both rural and urban areas in China, the majority of these providers are in rural areas Private health care facilities are often more accessible to rural communities in both cost and distance (Liu

et al., 2006) The effect of distance on provider choice is well documented in other developing countries and is addressed in policy primarily through contracting For instance, in Colombia, the Philippines, and Thailand, contributions into a social health insurance fund are used to purchase services that members want, from providers they

choose, and in close proximity to where they live (Hsaio and Shaw, 2007) The social health insurance fund within these countries allows for contracting with both public and private providers, holding them both accountable for quality and client satisfaction

While some literature has explored how patients choose a medical provider in rural and urban areas, this area of research remains incomplete Yip et al (1998) were the first to quantify the factors which determine patient choice of provider for the rural population at village, township, and county levels in China They find that patient choice is determined by insurance status, income, disease pattern, education, and age This same study is restricted in its approach, however, as it does not include private facilities where insurance is not accepted In addition, the study was only conducted in one county, Shunyi County near Beijing, a relatively rich subpopulation of rural China, and its conclusions, say the authors, “may not be generalizable to the rest of China.”

Liu et al (2006) find that although private services are not included in the

social insurance benefit package, these services continue to be used by low-middle income groups in rural areas The study reports that patients within these groups

choose private health care services on the basis of lower costs and higher quality of care This finding of utilization of private facilities by rural communities is not

congruent with other studies which show primarily middle-high income groups

choosing private facilities Private facilities offering lower costs than public facilities is also at odds with most other cases in developing countries (Wagstaff, 2007)

Health Insurance

There are signs that the Chinese government is changing its focus to address the particular needs of rural and urban residents in China The President, Hu Jintao, has recently unveiled a new campaign entitled the “Five Coordinations” which seeks to coordinate development between five distinct pairs: urban and rural areas; the economy and social programs; the environment and human beings; domestic and international demands This tall order may be only political propaganda, but the new policy does break from rigid traditional political thought in China and may foretell the emergence

of a new “softer” thinking in China (Wong et al., 2005)

One way the Chinese government plans to coordinate development in urban and rural areas is through its current plan of extending health care coverage to more than 80 percent of the country’s rural counties, a plan announced by Premier Wen Jiabao in March, 2007 (Xinhua, 2007) Under this voluntary system, a farmer

participant pays US$1.3 a year and the state, provincial, municipal and county

governments supplement this amount with US$5.2 This program allows only rural residents to use this joint fund in public hospitals

Other developing countries have implemented similar voluntary and subsidized health insurance programs for rural residents In 2003, Vietnam also introduced a program in which the poor were enrolled at taxpayer expense (Wagstaff, 2007) This new scheme resulted in patients switching from private providers to public providers Studies have documented similar effects in other developing countries For example,

Mexico and the Philippines have introduced voluntary tax-financed schemes where households, except for those in the poorest bracket, contribute according to their incomes (Knaul and Frenk, 2005; Obermann et al., 2006) Colombia has also

introduced subsidized schemes within its social health insurance (Escobar and

Panpolou, 2003) Data from Mexico suggests that implementing health insurance programs will increase utilization of public services, though the effect on private providers has not been explored In Colombia, insurance coverage did not increase public hospital utilization but was shown to increase preventative and ambulatory care

in public facilities

Wagstaff et al (2007) analyze the impact of China’s new insurance scheme which was implemented as a 2004 trial in a small population The paper reports the insurance scheme had several limitations in that many services were not covered and high deductibles made health care very costly As a result, the study finds having health insurance did not increase utilization of public facilities for the poorest bracket

of those surveyed This finding is consistent with the effects of a similar insurance scheme in Vietnam (Wagstaff, 2007)

In contrast, households with higher incomes did increase their utilization of public facilities in both China and Vietnam This finding may be explained by the moral hazard resulting from the insurance program and is consistent with findings of similar studies in Colombia, Mexico, and the Philippines Patients seek more care as a result of being covered, thus increasing public utilization Because private facilities outnumber public facilities in rural China, more research needs to be done on how the

insurance scheme and moral hazard affect overall utilization, including the area of private providers, especially for households in the lowest income quintile

The new insurance scheme in rural China has not been shown to reduce pocket spending (Wagstaff et al., 2007) This finding is consistent with experience in Vietnam, but remains at odds with the findings of similar studies in the Philippines, Colombia, and Mexico However, while the effects are not always large, the insurance

out-of-scheme in China is unique in that it may actually increase out-of-pocket spending

Lindelow and Wagner (2005), for example find that health shocks (serious illness and injury) may create a greater financial burden for those with insurance than those without insurance Similarly, Shi and Chen (1998) confirm reports of significantly higher medical expenses for the insured than for the uninsured with the same health problems and treatment outcomes Lindelow and Wagner suggest that moral hazard may be one factor explaining this increase Because patients may be more likely to see

a doctor when they have health insurance, out-of-pocket spending may also increase for those with health insurance

In addition, the increase may be attributable to physicians overcharging those with health insurance Li and Li (2002) indicate that the main cause of differences in medical expenses between the insured and uninsured is the cost of drugs Attributing the difference in cost to the effects of moral hazard either in the form of patient price-elasticity of demand or supplier-induced demand for those able to pay remains unclear (Eggleston et al., 2006) Several studies find that Chinese public health facilities burden patients with excessive drug prescriptions, high hospital expenditures, and

unnecessary medical treatments (Liu and Mills, 1999; Zhang et al., 2003; Zhan et al., 2004) Meng et al (2000) report that both public and private facilities in China

overcharge patients equally The effect of moral hazard on patients’ choice of provider remains little-researched

Another potential problem of the new insurance scheme is adverse selection Wang et al (2006) find that adverse selection may cause the new voluntary insurance schemes in China to be financially unstable They find that the sickest patients are enrolled in the rural voluntary insurance scheme, but the healthiest patients have opted out of the program While the effects on the private sector are not included in their research, this finding may support the notion that those with less severe illnesses choose private facilities while those with more serious conditions choose public

facilities As a result, the health care insurance reforms which China continues to choose may be unsustainable as those who enroll in the insurance scheme are likely to

be the most costly patients Since the study by Wang et al is limited to a very small segment of the rural population for a trial, more research needs to be done to support this finding In addition to difference in price, the perceived difference in quality between public and private clinics may also lead some patients to choose one facility over another

Because the Chinese government’s goal in expanding insurance is to increase the overall access to care, there needs to be more research on whether this goal is being achieved This paper seeks to fill this gap and find whether overall service delivery is

increasing as a result of insurance or whether it is only moving patients from private to public health care facilities

The Quality of Private Health Care in China

Due to the informal training of medical teams working in the private sector, some rural counties have issued a ban on private health care facilities, stating that the quality is too low or the cost is too high Zhang and Qiao (2002) find that in China’s rural township health centers, just 41% of health workers had graduated from high school, while 30% had only a primary school education Despite this lack of education, however, the quality of private health care providers has surprisingly not been shown

to be dramatically inferior to public health care providers Meng et al (2000) report that although the quality of care in private facilities is poor, there is no difference

between public and private clinics In addition, work by Lim et al (2002) find that patient satisfaction within private facilities is often higher than within public facilities Most studies confirm that banning private clinics is not a useful policy option, but rather policies that seek to improve quality and access to both public and private

facilities should be pursued (Lindelow and Wagner, 2006)

As there appears to be little difference in quality between private and public providers, this paper seeks to understand how health insurance may affect public

versus private provider utilization Cost is clearly an incentive for patients to move from private to public providers, but the goal of China’s health insurance is to expand overall coverage This area of research has been largely under-researched Such an

analysis will fill the gap between studies on the quality of private providers and studies

on insurance policy in China A better understanding of the effects of insurance on patient behavior may aid policymakers to find methods that will more effectively reach rural and low-income households

CONCEPTUAL MODEL

To analyze whether health insurance is accomplishing its goal of expanding access to health care, this paper compares the statistical outcome of health insurance on both public and private providers Therefore, the dependent variable in this analysis is

“provider choice” which represents the decision made by household respondents to choose a public or private health care provider The main independent variable of interest is health insurance

If patients are simply switching from private to public providers but using the same amount of health care services, then I expect the degree to which health insurance

is negatively associated with private provider choice to be equal in magnitude to the degree to which insurance is positively associated with public provider choice In this case, health care is not being created; it is simply transferring money from people with mostly good health to people with mostly poor health On the other hand, if the degree

to which health insurance is positively associated with public provider usage exceeds the degree to which it is negatively associated with private provider usage, then health insurance may be producing a net expansion of total health service provision

This paper seeks to explore these issues using the dataset by Lim et al (2002) These authors find that households with health insurance are more likely to choose private clinics than those households without insurance This pattern is particularly interesting because social insurance is not accepted at private clinics in China

Therefore, this paper first attempts to replicate Lim et al.’s finding of a positive

correlation between having insurance and seeking care at private facilities As this finding is counterintuitive and contrary to previous studies, this paper then analyzes the various independent variables which may explain this phenomenon I hypothesize that the correlation arises from systematic differences in the traits of households with and without insurance, rather than from the insurance itself Interacting these independent variables with health insurance reveals what types of people react more strongly to insurance in their decisions The characteristics I posit to explain as the pattern in provider choice include household income, health status, the particular Chinese

province, the view that public physicians overcharge patients, and the location of the household (rural or urban)

Households with higher incomes might be better able to afford private clinics’ fees, but they are also better able to afford transportation to distant public clinics The relationship between income and provider choice might therefore be more complex

Household income could also explain potential moral hazard effects generated from the insurance scheme Evidence suggests that those insured often have higher out-of-pocket costs than the uninsured due to either increased utilization or to the tendency of providers to given unnecessary care to the insured (Wagstaff et al., 2007)

If higher out-of-pocket spending is attributable to an increase in demand among the insured, then it is likely that that the insured will seek care at public providers This finding, however, is not consistent with the findings of Lim et al., who find the insured are more likely to visit private facilities than public facilities If the higher cost is attributable to public providers giving more unnecessary care to the insured, then there may be a tendency of those with insurance to switch to private providers to minimize this risk (Lim et al., 2002) Evidence of higher cost for the same care at public

facilities compared to private facilities would also be expected

Households with health insurance may also make the decision to visit a public

or private facility on the basis of their health status Patients may visit private facilities for minor health care issues and public facilities for more major issues Evidence of this adverse selection has been documented as being a potential problem for insurance schemes in rural China where the severely ill and more costly patients enroll while less costly patients opt out of the voluntary insurance scheme (Wang et al., 2006) Patients with chronic illnesses may also choose one facility over the other If adverse selection

is a problem, then those with health insurance who choose to visit a private facility may do so only for minor health concerns

Age, education, marital status, occupation, and gender could also affect choice

of health care facility More educated people may be more health-conscious and have more income, which may also influence provider choice Elderly patients who suffer from more serious illnesses may also be more likely to utilize health insurance

We must interpret any correlation between these traits and provider choice, however, bearing in mind that some such correlations can arise either because provider choice itself causes the characteristic in question, or because something associated with provider choice does, or because something about provider choice causes people to make different decisions about insurance Choosing providers who offer only provide lower-quality care, for example, could extend treatment and thereby decrease

household income from missed workdays Quality or availability of care, certainly related to provider choice, could even influence whether a person chooses to live in a rural or urban area; those in rural communities might move to urban areas where health care is more readily available The cost of care may also be affected by provider choice as clinics seek to become more competitive Such caveats are needed in the interpretation of this and all other retrospective evaluation

The study’s questionnaire was administered by interviewers to 3,600

households in three provinces, involving both rural and urban areas, between January and December 2001 The study also included a self-administered questionnaire survey

of 720 medical practitioners and 24 focus group sessions for patients and practitioners

in the three provinces Interviews with health officials, health care managers, and health care investors were also conducted during this time This paper, however, will only focus on the interviewer-administered questionnaire surveys given to the 3,600 households

The household questionnaires were administered to heads of household or the equivalent, defined as any person living in the house, male or female, aged 18 years or older, and a Chinese permanent resident Respondents were excluded if they were: below age 18; mentally or cognitively impaired; too sick or weak to answer questions; unable to answer questions because of other reasons; or unable to use or understand Mandarin Participation from households was strictly voluntary Invitations were sent via mail followed by home visits where participants were guaranteed full

50 households was conducted by the medical students who were also required to pass

an exam before officially administering the surveys University teaching staff

members were deployed as supervisors For quality control, supervisors reinvestigated 5% of all household surveys

The three provinces where the questionnaires were administered—Guangdong, Shanxi, and Sichuan—are geographically disparate and were purposefully chosen, based on their different stages of economic development and on the availability of suitable research collaborators in the provinces Selection of households within these provinces was determined by the use of multistage cluster sampling The sampling was conducted as follows

First, the health bureaus of the three provinces were asked to suggest urban cities and rural counties that they thought were “typical” of the province and which were also known to have private medical providers Additional criteria included the support from the local government and health department officials necessary to

conduct such a study The sample size of 600 households for each city or county was based on the goal of achieving a 95% confidence level for estimated mean values of responses After the counties and cities were chosen within the three provinces, two urban districts and two rural townships were then randomly selected within each

county and city In addition, within these districts and townships, two urban residential committees and two rural village committees were randomly selected Finally, the households within these committees constituted the cluster of households to be studied The target sample population was 150 households from the total households under the jurisdiction of each committee A total of 180 households were randomly chosen, with the last 30 households designated as “reserve households” for replacements if

necessary

ANALYSIS PLAN Dependent Variables

To analyze the effects of insurance on households choosing public or private facilities, I employ a logit regression model I first create indicator variables for use as dependent variables and attempt to establish the simple correlations between insurance coverage and public or proviate facility choice The definition of a “private provider”

in this dataset is based upon ownership All non-governmental institutions are defined

as “private.” “Public provider” is defined as a governmental institution Clinics and hospitals are analyzed both separately and combined I employ the model separately

on several dependent variables which describe choosing a public or private provider

To check for robustness, I include a dependent variable of patient preference to be seen

by a private doctor over a public one Testing these different dependent variables is necessary to find common trends and bring about more robust results

Because only 5% percent of the observations sought treatment at private

hospitals whereas 53% sought treatment at public hospitals, I separate hospitals from clinics I then compare these trends among the other dependent variables in the dataset which are reported within Table 6 and discussed in more detail below

For the creation of the first set of dependent dummy variables, household answers to the following question are used: “Where did you see the doctor on your last visit?” Based on respondent answers, the following six indicator variables are created:

1) “Private clinic on last visit:” If the respondent answered “private clinic”, the dependent variable takes a value of 1 (otherwise, 0)

2) “Public clinic on last visit:” If the respondent answered “public clinic,” the dependent variable takes a value of 1 (otherwise, 0)

3) “Private hospital on last visit:” If the respondent answered “private hospital,” the dependent variable takes a value of 1 (otherwise, 0)

4) “Public hospital on last visit:” If the respondent answered “public hospital,” the dependent variable takes a value of 1 (otherwise, 0)

5) “Private facility on last visit:” This variable combines both private clinics and private hospitals to measure overall health insurance effects on private facility utilization If the respondent answered either “private clinic” or “private

hospital,” the dependent variable takes a value of 1 (otherwise, 0)

6) “Public facility on last visit:” This variable combines both public clinics and public hospitals to measure overall health insurance effects on public facility utilization If the respondent answered either “public clinic” or “public

hospital,” the dependent variable takes a value of 1 (otherwise, 0)

For the creation of the second set of six dependent dummy variables, household answers to the following questions are used

1) “Private clinic in the last 12 months:” If the respondent answered “yes” to the question, “In the last 12 months, have you been admitted to a private clinic?” the dependent variable takes a value of 1 (otherwise, 0)

2) “Public clinic in the last 12 months:” If the respondent answered “yes” to the question, “In the last 12 months, have you been admitted to a public clinic?” the dependent variable takes a value of 1 (otherwise, 0)

3) “Private hospital in the last 12 months:” If the respondent answered “yes” to the question, “In the last 12 months, have you been admitted to a private

hospital?” the dependent variable takes a value of 1 (otherwise, 0)

4) “Public hospital in the last 12 months:” If the respondent answered “yes” to the question, “In the last 12 months, have you been admitted to a public hospital?” the dependent variable takes a value of 1 (otherwise, 0)

5) “Private facility in the last 12 months:” This dependent variable combines private clinics and private hospitals If a respondent answered “yes” to the questions in 1 or 3 above, then the dependent variable takes a value of 1

(otherwise, 0)

6) “Public facility in the last 12 months:” This dependent variable combines public clinics and public hospitals If a respondent answered “yes” to the questions in 2 or 4 above, then the dependent variable takes a value of 1

To check for robustness of these results, I include a final dependent variable on patient preference using a question from the dataset, which asks respondents their degree of agreement with the statement, “When I’m sick, I prefer to be seen by a private doctor than a public doctor.” This dependent variable takes a value of 1 if the respondent agrees or strongly agrees and 0 otherwise Understanding patient

preferences is important to compare with actual patient behavior, as patients may not necessarily be going to the facility of their preference because a particular facility is not easily accessible Therefore, demand for a particular facility may be higher than the supply available and using patient preference as a metric for demand is useful in interpreting the results Comparing the results of these regressions among the various dependent variables adds robustness and confirmation of particular trends on the effects of health insurance

Independent Variables

The independent variables throughout this paper are described below

Health Insurance—This is an indicator variable for health insurance (X1) where 1= household has health insurance (insurance here includes all forms of insurance: state health insurance, co-operated medical services, and company health insurance as

each of these insurance programs are not accepted at private facilities) X1 takes the value of 0 if the household does not have health insurance.1

Urban/Rural and Province—The location of the household and the category of the household being either urban or rural may affect a household’s decision to visit a public or private facility From the dataset, I measure this effect using the variables: X2= Urban/rural, where 1 = urban, else = 0 The definition of being urban or rural is designated by the health bureaus within the three Chinese provinces

X3= Guangdong Province, where 1= lives in Guangdong province, else = 0

X4= Sichuan Province, where 1=lives in Sichuan province, else = 0

The omitted base group for X3 and X4 is Shanxi Province

Physician overcharging patient—To measure potential supply-side moral hazard effects of physicians overcharging patients with health insurance in public facilities, I would ideally like to know the exact amount that patients are charged for services in public facilities and compare this amount with what patients are charged in private facilities As this information is unattainable within the dataset, a proxy on patient opinions of being overcharged is employed This variable (X5) is taken from a

question in the dataset which states: “Doctors in public practice tend to over prescribe

medicines for patients in the purpose of earning money.” Variable X5 takes the value

1 State health insurance is defined as government-issued health insurance outside of rural areas operated Medical Services is the social health insurance issued by the government particularly for rural areas Company Health Insurance is private health insurance issued from an employer

Co-of 1 if the household reported that they “strongly agree” or “agree” with the statement

Variable X5 takes the value of 0 for any other response

Health Status—To analyze the potential effects of adverse selection, including the patient’s health status in the model is necessary Patients with health insurance may make the decision to visit a public or private provider on the basis of how sick they are, possibly choosing private facilities for minor health care issues and public facilities for more major issues Patients who suffer from chronic illnesses may also choose one facility over the other To measure this effect, I use a question from the survey which asks respondents to give information on their present health status From this

information, I create the variable X6 which will take the value of 1 if the respondent reports their present health status as being either “fair”, “good”, or “very good.” Variable X6 will take the value of 0 for any other answer

Income—Households may make the decision to choose a private facility on the basis of their income From the dataset, I measure this effect using the variable:

• X7= Annual combined household income where income is 2000-4999 yuan (USD$280-695)

• X8= Annual combined household income where income is 5000- 9999 yuan (USD$696-1,390)

• X9= Annual combined household income where income is 10000-19999 yuan (USD$1,391-2,780)

• X10= Annual combined household income where income is 20000 yuan and above (USD$2,781+)

Annual household income less than 2000 yuan is omitted as a variable and included in the baseline group

Demographics—Additional demographic factors that could influence provider choice are also incorporated in the model These variables include:

• X11= Age

• X12= Male where male=1, female=0

• X13= Education where primary school is the highest level of education

• The omitted baseline group for Education is “No education.”

• X16= Marital status where married =1, else =0

• X17= Occupation where government officer = 1, else=0

• X18= Occupation where Manager/executive=1, else = 0

• X19= Occupation where Clerk/serviceman=1, else = 0

• X20=Occupation where self-employed=1, else=0

• X21=Occupation where farmer=1, else=0

• X22=Occupation where student/part time/others= 1, else = 0

• X23=Occupation where retired = 1, else = 0

The omitted baseline group for occupation is “Unemployed.”

Thus the model I use for the results in Tables 7-10 is:

Y= β0 + β1X1 + β2X2 + β3X3 + β4X4 + β5X5 + β6X6 + β7X7 + β8X8 + β9X9 + β10X10 + β11X11 + β12X12 + β13X13 + β14X14 + β15X15 + β16X16 + β17X17 + β18X28 + β19X29 + β20X20 + β21X21 + β22X22 + β23X23

By interacting these independent variables with health insurance, this study is able to estimate whether the correlation between insurance and provider choice depends on another independent variable Some groups may be more sensitive to health care coverage than other groups in determining whether to go to a public or private health care facility For example, if provider choice is regressed on the

independent variables of “insurance,” “urban,” and “insurance x urban,” then the coefficient on the interaction term will describe the degree to which the correlation between insurance and provider choice depends on whether or not the household lives

in an urban area This interaction term could also describe the degree to which the correlation between living in an urban area and choosing a private provider depends on whether or not the household has health insurance If the covariate on the interaction term has a large positive sign in the public regressions and a small negative sign in the private regressions, then insurance particularly affects health care provision among people in urban areas

Similarly, if provider choice is regressed on “insurance”, “Sichuan province”,

“Guangdong province,” “insurance x Sichuan province,” and “insurance x Guangdong province,” then the value of the coefficient on the two interaction terms expresses the degree to which the relationship between provider choice and insurance depends upon the province the household lives in, or the degree to which the relationship between provider choice and province depends on insurance Comparing the coefficients of the interaction terms between the public and private regressions reveals whether particular provinces are more or less inclined to visit public or private facilities

If provider choice is regressed on “insurance,” “physician overcharging the patient in public facilities,” and “insurance x physician overcharging the patient in public facilities,” then the coefficient on the interaction term describes the degree to which the correlation between insurance and provider choice depends on whether or not the household believes that public facilities overcharge patients, or vice versa Comparing the coefficients on the interaction terms between the public and private regressions reveals whether those who believe public facilities overcharge patients are more or less inclined to visit public or private facilities

Having health insurance is also expected to have different effects on choosing a private provider among households in different income brackets, good and poor health statuses, and households with high or low satisfaction with the convenience of the location of the provider Therefore, interaction terms of insurance and these

independent variables are also included in the model Comparing the coefficients of the interaction terms between the public and private regressions reveals whether these particular segments of the sample are particularly sensitive to the expansion or health care coverage or switching from public to private clinics

Thus, the model I use to generate the results presented in Tables 11-13 can be expressed as:

Y= β0 + β1X1 + β2X2 + β3X3 + β4X4 + β5X5 + β6X6 + β7X7 + β8X8 + β9X9 + β10X10 + β11X11 + β12X12 + β13X13 + β14X14 + β15X15 + β16X16 + β17X17 + β18X28 + β19X29 + β20X20 + β21X21 + β22X22 + β23X23 + β24(X1*X2) +

β25(X1*X3) + β26(X1*X4) + β27(X1*X5) + β28(X1*X6) + β29(X1*X7) +

β30(X1*X8) + β31(X1*X9) + β32(X1*X10) + β33(X1*X11) + β34(X1*X12) + β35(X1*X13)

Discussion

Any estimate of the true partial correlation between the regressors and provider choice will be biased if important variables correlated with both of them are omitted from the regression model An important additional variable that may influence choice

in provider is the cost of treatment, but costs are not observed in the survey data in a