The Impact of the New Rural Cooperative Medical
Scheme on Activities and Financing of Township Hospitals
in Weifang, China
Xiaoxian HUANG, Aurore PELISSIER1, Martine AUDIBERT, Jacky MATHONNAT
Centre d’Etudes et de Recherches sur le Développement International (CERDI),
University of Auvergne
65 bd François-Mitterrand, 63000 Clermont-Ferrand, France
Abstract:
Since 2003, the New Rural Cooperative Medical Scheme is gradually implemented in China, in
order to increase access of the poor to health services, reduce out-of-pocket expenditures and avoid
catastrophic health expenditures. Township hospitals are the first level health facilities in Chinese
referral system of medical services. The paper deals with an impact analysis of the NRCMS on
activities, efficiency and financial structure of township hospitals in Weifang prefecture, Shandong
Province. Twenty four township hospitals and nine years (2000-2008) composed the dataset. The
NRCMS reform was implemented gradually in townships, between 2003 and 2006.
Results from the estimation of the fixed effects model show that NRCMS has a significant impact
on activities and financial structure of township hospitals. Both the number of outpatient visit and
discharged patient increased with the implementation of the reform, which means that the insurance
scheme positively impact utilization of health services. Also, the average length of hospital stay
(ALOS) and the bed occupancy ratio improved. NRCMS decreased average length of stay at hospital
and increased the bed occupancy rate. A better utilization of township hospital inputs is permitted by
the insurance reform. Concerning financial structure, the share of business income generated by drug
selling increased with the implementation of the reform.
Two sources of heterogeneity are checked. Impact of the reform is different over time and
according to the economic status of the township. In a general way, township hospitals located in poor
townships experienced a greater impact than non poor ones. The impact of reform seems enlarged one
year after the implementation for the case of hospitalization.
JEL Classification: O12, I11, I38
KEY WORDS: China, New Rural Cooperative Medical Scheme, Impact analysis, Township
Hospitals.
1 Corresponding author: PELISSIER Aurore
Contact address: CERDI. 65 boulevard F. Mitterrand. 63000 Clermont-Ferrand.
Email address: [email protected]
Contact phone number: +33 618925468
This study is part of a research program. This program benefits of supports from University of Auvergne and the
Embassy of France in Beijing.
WORK IN PROGRESS
PRELIMINARY DRAFT
2
1 Introduction
During the years of 1960s to 1970s, Chinese government had constructed an efficient rural health
system which successfully ensuring the access to basic medical services for 85% of the Chinese rural
population with extremely low cost2 (Hsiao, 1995; Xueshang et al, 1995; World bank 1997). This
system was composed of a public health service delivery system and a community-based health
insurance system.
Rural health delivery system has a three-tier structure, composed, from lower to upper level, of
village health stations, township hospitals, and county or above-level hospitals. Township hospitals
have an essential role in this delivery system, as they are the linkage between health stations and
county hospitals. On the one hand, they supervise the quality of services delivered at village health
stations and offer technique backup to the latter and are able to provide general surgery care. On the
other hand, they play the role of gate keeper who filters needed patients to upper-level hospitals. In
addition, they are also in charge of the organization and delivery of preventive care to the rural
population. They are therefore the main provider of primary healthcare in rural areas (Hillier and Shen,
1996). To guarantee that the medical services are affordable to the population, service prices were set
below costs (World Bank, 1997). Therefore township hospitals need local and central government
subsidies to cope with a potential deficit. In another words, the sustainability of township hospitals
depends on the public financing capacity.
The rural health insurance system, conventionally referred to as “Rural Cooperative Medical
System” (RCMS), was a community-based health insurance system, organized by agricultural
collectivity. Each collectivity was charged to ensure the accessibility of basic medical services to all
the members given the budget constraint of insurance pool. The community was then the footstone of
the RCMS.
Since 1978, China has steered in an economic transition process. Although the reforms
undertaken brought great economic gains, they caused important perturbation in Chinese health
system (Hillier and Shen, 1996; Eggleston et al, 2008). The main changes concerned the fiscal
decentralization and the replacement of rural production collectivity by household production
responsibility system. The former destabilized the financial source of township hospitals, and the latter
ruined the financial basis of RCMS. On the one hand, with the decentralization, the charges for social
welfare shifted to the local government. This directly caused the shortage of public resource for health
sector in some regions. First, as the local development level varied across regions, the government’s
capacity to finance health sector differed accordingly. Second, with the process of privatization, both
the central and local governments’ income, measured in share of total GDP, decreased, implying a
2 Household contributions represented only 0.5%-2% of the peasant family income (Asian Development Bank, 2002).
3
general weakening of public financial capacity. Lastly, the economic reform evoked urgent needs for
public investment in economic sector, such as infrastructure. This caused competition for public
financing inside of public sector and reduced government’s willingness to increase public health
expenditure. On the other hand, due to the shortage of public financing, township hospitals, as other
health facilities, are encouraged to make up the budget deficit by increasing business income (Liu, Xu
and Wang, 1996). This fundamentally shifted the orientation of hospitals’ behaviors from social
welfare needs to economic interests. The supply of expensive curative activities increased at the
detriment of preventive and basic curative healthcare (World Bank, 1997). On the other hand, as
township hospitals are less well equipped than county hospitals, the market competition also put many
township hospitals into great financial hardship. As long as services were affordable, patients prefer to
consult at county or above level hospitals for better quality of services, even if they were more
expensive. The failure of township hospitals blocked the referral system and disconnected health
stations from upper-level hospitals. Both considerably reduced the effectiveness and efficiency of
public health delivery system. Finally, with the implementation of household production responsibility
system in rural areas (1981), agricultural communities gradually disappeared. As it was the basic
financial source of RMCS, it directly induced the collapse of rural health insurance system (Hsiao,
1984). At the beginning of the 1990s, less than 10% of the rural population was covered by a medical
insurance system and out-of-pocket payment sharply increased (World Bank, 1997). The financial
obstacle became one of the main causes of poverty and poor health. According to the National Health
Survey (2003), half of the population didn’t go consulting in case of disease (World Bank, 2009).
Three quarters of the rural population that needed hospitalization couldn’t have it due to the financial
hardship (World Bank, 2009).
In such circumstances, the reduction of the efficiency of primary health system drew more and
more government’s attention. Since 2003, the government carried out a series of reforms aiming at re-
establishing rural health insurance system and increasing the financial accessibility of health service to
rural population. This new system is called “New rural cooperative medical system” (NRCMS) to
distinguish from the former one. The coverage of NRCMS enlarged quickly (Table 1). The objective
is to achieve universal coverage by 2013 (Anonymous, 2009b in Yip and Hsiao, 2009). This system is
managed at the county level, but two directives are national: the engagement is voluntary and the main
goal is to protect households from impoverishment due to catastrophic health expenses (Central
Committee of CPC, 2002; Yip and Hsiao, 2008, 2009; You and Kobayashi, 2009). Another important
consideration is to use economic incentives, such as reimbursement rules or provider payment, to
divert the population back to lower level health facilities and encourage the township hospitals to offer
more cost-efficient services (World Bank, 2009).
The importance of taking the health provider aspect into consideration when evaluating the
impact of insurance reform is widely recognized. A study found that in Weihai, people’s satisfaction
4
with insurance scheme is at least partly linked to the supplier side, such as receiving health check-up
and having shorter average length of hospital stay (ALOS) (Liu, Xu et al, 2008). In order to evaluate
incidence of NRCMS, the World Bank and the Chinese Ministry of Health surveyed 89 counties (in
which 27 were questioned in a more detailed way). Wagstaff, Lindelow et al (2009) exploited this
database to conduct an impact analysis. Surveys and estimates revealed several interesting futures.
First, even if insurance coverage is high, the problem of adverse selection persists. Second, despite of
this, they found that NRCMS has a positive impact on outpatient, inpatient visits to the township
hospitals and on bed occupancy rate, but not on out-of-pocket spending and ALOS.
This paper investigates the effect of NRCMS on the situation of a sample of township hospitals of
Weifang prefecture. The main question to answer is whether the implementation of insurance
influences the medical activities and financial structure of township hospitals. Can insurance reform be
an opportunity to re-establish the township hospital’s reputation and thus contribute to the good
function of referral system? These questions are essential for the harmony of insurance reform with
other reforms in health system, such as hospital and pharmaceutical reforms.
In order to take into account incentives which drive health suppliers’ behavior, an econometric
model is developed to identify the effect of the introduction of the insurance system on township
hospitals’ activities and financing. A unique and original dataset was collected in the Weifang
prefecture of the Shandong province in China. It contains rich information on a group of randomly
chosen township hospitals in the province. The data covers a period of 9 years (2000-2008).
Econometric analysis exploits the schedule of NRCMS implementation, beginning in 2003 and
completed in 2006 (in our sample). Identification tests were performed to valid the econometric
model, including tests for parallel trend hypothesis, selection bias and issues on migration. Then, the
heterogeneity of impact is investigated.
Section 2 presents the context in which NRCMS was implemented in Weifang prefecture. Section
3 is devoted to the development of empirical model of township hospitals behavior. Section 4 exploits
the data and conducts econometric analysis. Results are shown in Section 5. Finally, Section 6 gives
out the conclusions and discussions.
2 Background of NRCMS reform in Weifang prefecture
Weifang is a city-prefecture of Shandong province locating in the northeast of China. It is
composed of 16 county-level administrative divisions3. To simplify the appellation, they are jointly
referred to as “county” in the following study. These counties have jurisdiction over 148 townships (or
3 They are composed of four districts, six cities, two counties, three development zones and one economic zone in reservoir
area.
5
districts) and 9285 villages. The total population is 8.5 million, in which 6.22 million are rural
population, accounting for 73% of total. In terms of health infrastructures, there are 6384 health
facilities, including 61 county-or-above-level hospitals and 194 township hospitals (personal
communication from Weifang Health Bureau). In our sample, there is one township hospital per
township.
The reform proceeds in rural areas in a gradual way since 2003. At the initial stage, the Weifang
government has chosen some pilots counties which then chose a number of townships for applying the
reform. New counties and townships were involved into the reform in the following years. Until 2006,
all the municipalities and counties in Weifang are eligible to the insurance reform. Universal coverage
of rural population is obtained by 2009 (personal communication from Weifang Health Bureau).
Several features characterize the design of the insurance system. First, insurance is pooled at
county level and managed by the government. The prefecture government set NRCMS bureau to
guarantee the overall planning of reform. The objective is to facilitate the coordination and
cooperation of county NRCMS funds. County governments establish special NRCMS committees to
carry out insurance implementation. This is done with the help of NRCMS agencies at township,
which are in charge of the management and reimbursement of insurance inside of township. In
parallel, a supervision committee is also organized at county level to supervise and audit insurance
funds. Second, the insurance fund is collected from three agents: the individual, the local and central
governments. Famers participate to the insurance scheme by household and on a voluntary basis.
Contribution was 10 Yuans per person in 2001 and gradually rose to 25 Yuans by 2009.
Correspondingly, local and central governments’ subsidies increased from 20 Yuans to 40 Yuans per
person respectively. The total contribution attained 105 Yuans per person by 2009. Until 2006,
farmers’ contributions for the next year are collected in May or June of the present year. However, in
2006, the time of collection shifted to November. Once individual’s contributions are collected, the
subsidies of local and central governments are allocated in accordance with the number of participates.
At the initial stage, funds are divided into two accounts: a personal medical saving account (MSA) for
outpatient spending and a mutual assistant fund for expenditures on more complicated treatments
(such as the treatment for chronic diseases or hospitalization). However, since 2009, to promote the
recycling of funds, all are pooled at county level and then divided into outpatient funds and
hospitalization funds. The first one accounts for about 30% of total funds while the second one
represents 67% of the total. The rest 3% is set as risk fund. Third, the reimbursement process is a little
bit complex. Deductibles, caps and co-insurance ratios are therefore specified for expenditures at
different health facilities. As treatment costs at lower level hospitals are cheaper than at higher level
ones, the insurance is much more generous for the use of health service at lower-level hospitals.
Precisely, the deductible is set at 100 to 300 Yuans for the consultation at first-level hospital
(corresponding to township hospital), 400 to 600 Yuans at second-level hospitals, and 1500 Yuans at
6
third-level hospitals. The cap is 40,000 Yuans per person per year whatever health facilities. The co-
insurance for hospitalization is 60 to 70% at first-level hospital. For higher level hospitals, co-
insurance rate changes in function of total expenditure and is generally less than 60% if expenditures
are lower than 5000 Yuans. The reimbursement for outpatient is confined to the consultation at first-
level hospitals or health stations and reimbursement ratio is set around 20% (personal communication
Weifang Health Bureau). In 2009, the share of reimbursement in total health expenditure in Weifang
was around 35%. In order to incite farmers to join the insurance scheme, an annual comprehensive
physical exam is also organized for the insured whose have not enjoyed the insurance reimbursement
during the covered year. An office of NRCMS is set inside of the township hospital to perform the
reimbursement on the spot. Before 2008, the insured need to seek the medical consultation at
designated hospitals. However, great discrepancy in the quality of services offered by designated and
non designated hospitals in some townships makes households bypass the former at the expense of
abandoning the insurance reimbursement. In return, they are reluctant to continue the adherence due to
the lack of benefits obtained from the insurance (Sepehri et al., 2009). Since 2008, the rule of
designated hospitals is released. The patient can choose any hospital inside of the county and get the
reimbursement.
There are several objectives lying behind the design of the reform. First, the activities of township
hospitals are expected to increase, because the insurance reduces patients’ financial obstacles. Second,
insurance can also ameliorate the financial structure of township hospitals by modifying economic
incentives for the provider. These objectives will be assessed by the following empirical study.
3 Econometric model
In order to estimate the NRCMS impact, we use a quasi-experimental and spread method (Aker,
2008; Galiani et al, 2008). This permits to exploit the panel dataset and the gradual implementation of
the health insurance reform over time. The following model is estimated:
[1]
where Yijt are the activities and financing variables for township hospital i in county j at time t. As
there is only one township hospital per township, we use “i” to represent either township hospital or
township. Pijt is a dummy variable equals to one if in year t NRCMS is applied in township i of county
j, and 0 otherwise. The coefficient of Pijt, δ, is the estimator of interest. It represents the estimated
average effect of the NRCMS implementation on hospital outcomes. Xijt is a set of exogenous
regressors varying across time and space (townships hospitals and townships characteristics) which
may affect the outcome Yijt. Township hospitals fixed-effects (ui) and year fixed-effects (vt) are also
included in the specification. The former controls for time-invariant individual characteristics, whereas
7
the latter captures township hospitals common characteristics varying over years. εijt is the error term.
Equation 1 is estimated in fixed effects with standard errors corrected for heteroskedasticity and serial
correlation, respectively by Huber-White and Newey-West methods. To detected AR(1) serial
correlation, Wooldrige test is employed.
4 Empirical analysis
4.1 Data and measurement
This paper uses a unique dataset, building from a survey conducted in collaboration with the
Studies and Researches Center in International Development (CERDI) of Auvergne University, the
Weifang Health Office, and the Medical University of Weifang. It is a longitudinal survey which
covers the period from 2000 to 2008 for Weifang prefecture. A sample of 24 township hospitals from
six counties was randomly selected among the total 148 township hospitals in the prefecture. General
information about the characteristics (size of the population, per capita rural income, insurance
coverage rate, etc.) of counties, townships and NRCMS, were collected. At the township hospital
level, data concern hospital resources (beds, equipment, staff, financing, etc.), activity indicators
(outpatients, discharged patients, immunizations, etc.), and NRCMS reform (date of the entry,
coverage rate of patients, etc.). Sources of data include registers of township hospitals, Statistical and
Finance Offices of townships and counties, as well as interviews with directors and principal officials
of each township hospital.
Rural household’s participation rate is generally high and increases rapidly over time. In 2006,
the coverage of insurance attained more than 90% of the sample population. Three reasons may be
advanced. First, the contribution to the insurance is affordable comparing to the rural income. Mao
(2006) investigates 282 pilot NRCMS counties and finds that the premium accounts for about 0.44 to
0.57% of overall farmer income according the economic development of the region (Mao, 2006).
Second, the government gives important subsidies to ensure that the level of reimbursement is
significant, especially for important health expenditure. Third, the local government considers the
participation rate as an indicator for the performance of their governance, and thus does great efforts to
mobilize the population.
Five outcome variables are selected. Medical activities are measured by the number of outpatient
visits, the number of discharged patients4, the ALOS and the bed occupancy ratio. Changes in
financial structure are captured by the share of drug selling in business income. The three first ones are
transformed on logarithm, whereas bed occupancy ratio and share of drug income are in level.
4 It is a conventional indicator to measure the volume of inpatient.
8
4.2 Descriptive Statistics
Tables 3a and 3b present descriptive statistics of the main analysis variables. On average, the size
of townships where the sample township hospitals are located is 13,603 households with 92% of rural
population. The farmer net income per capita amounts 4581 Yuans5. Between 2000 and 2008, the
number of households and the average net income per capita increased, but the share of rural
population remained quite stable. Concerning health service supply, the number of village health
stations has increased by 6% per year during the period. The capacity of township hospitals measured
by the number of beds increases 19% per year.
In terms of the volume of hospital activities, three kinds of activities need to be distinguished:
outpatient visits, discharged patients,6 and vaccinations. Outpatient visits turn out to be the main
activities of township hospitals. In number of treated cases, the outpatient visits account for 74% of all
the three activities. Township hospitals activities increased significantly since the reform. All the three
indicators have doubled between 2003 and 2008, whereas they stayed quite stable during the three
years preceding the reform.
In terms of the performance of hospitals, two indicators are chosen: the bed occupancy and the
average length of stay (ALOS). There are some proofs on the amelioration of township hospital’s
performance. The bed occupancy ratio is almost doubled between 2003 and 2008 (from 38% to 61%),
while it is slightly declined (from 42% to 38%) in the pre-reform period (2000-2003). As far as ALOS
is concerned, patients stay five days at township hospitals. Over the sample period, ALOS is
somewhat reduced in the reform period but stay stable after reform.
In terms of the treatment capacity of hospitals, between 2003 and 2008, the number of beds
increased strongly (from 36 to 59 beds per hospital), the number of professionals increased from 49 to
67. Nevertheless, the qualification structure of health staff remains similar. Junior professionals are the
most important component, accounting for more than 80% of total.
Business income of hospitals has tripled between 2003 and 2008, while it stagnated during the
pre-reform period (2000-2003). A great part of this increase comes from the selling of drugs, which
was multiplied by four. Subsidies follow similar evolution. The total subsidize has more than doubled
between 2003 and 2008. Nevertheless, the business income remains the most important financial,
about 92% of total.
High profit from drug selling can be captured by comparing the drug income with its expenditure.
The former is not only more important in absolute term than the latter, but also more rapid in its
growth.
5 All the monetary terms are in 2000 constant prices. 6 Commonly used measure for inpatient activities
9
To conclude with insurance scheme, the average of reimbursement is much higher in 2008
comparing to that in 2003. The reimbursement for outpatient and inpatient increased by 10 times and 5
times respectively.
4.3 Identification
In order to obtain an unbiased, consistent and asymptotically normal estimator on program effect
based on equation [1], several assumptions need to be verified: i) the exogeneity of the reform and the
order in which the reform was implemented, ii) the absence of migration.
4.3.1 The exogeneity of the reform
The model requires the assignment of program exogenous across space and over time. NRCMS is
a national policy which is expected to be adapted in all rural areas in China. Therefore the potential
endogeneity comes from rather the date they enter into the reform than their eligibility. There are two
sources of selection bias: reverse causality and omitted variables. Reverse causality refers to the
situation where the outcome of township hospitals influence the order that the NRCMS is
implemented in corresponding township. Hospitals with more activities, for example, may be better
organized and thus have higher efficiency. At the initial stage of the reform, the authority may prefer
to choose the regions where hospitals have better performance. As consequence, regions where the
hospitals have more activities have more chance to be chosen in first. It may also be the case that the
variation in hospital outcome influences the chance that the township should be chosen into the
program. If, for instant, the township hospital has encountered rapid decrease of patients in preceding
years of the reform, the township may be more motivated to participate to the program in hoping that
the program should bring more patients. In these cases, the date of introduction of the reform is not
independent from the outcome of the hospital. Regarding to the problem of omitted variables, it is
possible that variables which vary across time and township hospitals (and so are not captured by the
fixed-effects estimator) have effects on both outcomes variables and program implementation but are
not introduced into the control variable vector. The income of rural household, for example, is a
variable that vary across townships and years. It can be correlated with township hospital activities, or
financing, because this variable reflects the capacity of household to pay for health care services and
also the living standard of the township. But it can also be correlated with the interest variable, Pijt,
because poorer township can have implemented the reform in first and the richer in last. This problem
can be solved by adding the omitted variables into the regression (in the vector Xijt).
Three strategies of test are adopted to verify the existence of reverse causality. The first is to test
the parallel trend assumption (PTA). It states that the intrinsic evolution of the outcomes of township
hospitals in the absence of reform should be independent of NRCMS implementation. One
conventional way to test the existence of PTA is to compare the pre-intervention trends of hospital
10
outcomes between the regions which enter into the program at different time (Barham, 2005; Aker,
2008; Wagstaff and Moreno-Serra, 2009). The hypothesis is that the pre-intervention trends of
township hospital outcomes would be the same in the post-intervention period if the program were not
introduced. This will be the case if, other things equal; the pre-intervention trends are not significantly
different among the regions with different entry dates to the NRCMS. Inspired by Barham (2005), the
test of PTA is conducted with the following specification:
[2]
t=2001-2002 (pre-intervention period, 2000 is drop because it is the reference period) and k=2004-
2006 (period after the first intervention in 2003). NRCMSk equals to one if the township hospital
enters in the reform during year k, 0 otherwise; dyeart are year dummies which captures the period
specific effect of 2002 and 2001. θtk represents, at year t, the additional period effect hold by the
regions which enter into NRCMS program in year k. If θtk are not significantly different from zero,
PTA is assumed and pre-intervention trends can be considered similar between township hospitals that
applied the reform at various years. Results are summarized in Table 4. In all cases, the test supports
the assumption that pre-intervention trends of outcomes are similar across groups which entered at
different time in the reform.
The second strategy is adapted to test whether the NRCMS implementation decision is correlated
with the variation of outcome variables during the preceding years of reform. The test is performed on
a pre-reform period sample. For each township, it consists of the data for the three years before the
implementation of the NRCSM7. The following specification is used to test the assumption (Gruber
and Hanratty, 1995; De Janvry, McIntosh and Sadoulet, 2009):
[3]
where year-l,ijt takes the value 1 if at year t, the reform will be implemented in l years for township i in
county j, and 0 otherwise. l = 1, 2 or 3. (year-3, ijt is drop because it is the reference period). County and
year fixed effects are included, noted as wj and vt respectively8. κl are the coefficients of interest. If κl
are statistically significant, it suggests that the entrance dates of township hospitals are subject to the
evolution of outcomes during the three years before the implementation of the NRCMS. The results
are shown in Table 5. None of the coefficients are significant, which implies that the date of NRCMS
placement is not driven by evolution of outcome trends during the three years before the
implementation of the reform. In another words, entry order is exogenous.
7 As the dataset begins in 2000 and the reform begins in 2003, for townships who first implemented NRCMS in 2003, the
information on the period before the implementation of reform is available for maximum three years. In order to keep the
same quantity of information for all the townships whatever their entry dates of reform, we choose the three years before the
implementation to construct pre-reform dataset. 8 County fixed effects rather than township ones are included into the model due to the small sample size of the sample.
11
Third, reverse causality is checked with the test proposed by Gruber and Hanratty (1995) and also
applied in the study of Wagstaff and Moreno-Serra (2009). This problem can be captured by adding to
the model a dummy variable reflecting the implementation of the reform the next year. The
specification is as following:
[4]
There is no problem of reverse causality when coefficient on Pijt+1, noted δ’ is not statistically
significant. Coefficient on the added variable (participation in the next year) is not significant in all
regressions9.
In last, omitted time-varying variables can be identified by comparing the means of a
comprehensive list of characteristics (at township and township hospital levels) between townships
hospitals chosen for the first year of implementation of the health insurance reform and townships that
entered into the insurance program later on (Barham, 2005; Aker, 2008). According to the strategy of
Aker (2008), the differences in means of covariates between townships entered in the reform in 2003
and those which entered after are calculated. Then, they are divided by standard deviation of the 2003
group. Covariates with score higher than 0.25 (in absolute value) are considered to be significantly
different, and thus will be entered into the model (Imbens and Wooldridge, 2007; Aker, 2008). Tests
of means are performed on 2002 (the preceding year of the NRCMS reform). According to Table 6,
six covariates are to be included: the number of households, the rural net income per capita, number of
healthcare professionals and that of operational beds, and the subsidies (allocated by county and
government).
4.3.2 Migration
The last issue which can potentially biased the result concerns the problem of selective migration
(Galiani, Gertler and Schargrodsky, 2008). The change of outcome may issues from the change of
characteristic of the target population due to the implementation of the program, rather than the effect
of reform itself. In our sample, different townships enter into the program at different time. Therefore,
at a given period, residents in townships which are not yet covered by insurance may want to go to
townships that are covered to benefit from the system. It will bias results if the immigrates are
significantly different from the local population. If, for instance, they are sicker than local people and
therefore have more important needs for medical services, the increase of medical activities in the
insured township hospital will not be due to the reform but rather to the characteristic of the
immigrated patient. This is not the case in this sample, because the beneficiary status depends on the
resident place of the population. People in the township which is not covered by NRMCS will not be
9 P_value on test variable is not given.
12
able to benefit from the insurance even if they seek medical consultation at covered hospitals. Another
possibility is that people move into the covered region. However, in China, the residence status is
generally linked with the birth place. It is costly to change the living place, even if possible.
4.4 NRCMS impact
Table 7 lists the results of model 1. The impact of the NRCMS is significant for all outcomes10
.
Concerning activities variables, the insurance system has a positive impact on both the number of
outpatient visits, that of discharged patients and the bed occupancy ratio. When a township enters into
the reform, its township hospital observes an increase of the number of outpatient visits of 17%, that
of discharged patients by 66%11
and that of the bed occupancy ratio by 7.5 percentage points. The
observed higher impact on discharged patient confirms to the expectation, because the design of the
reform puts the accent on the coverage of hospitalization fees. These results indicate that the insurance
considerably increases individual’s access to medical treatment associated with important spending,
such as for hospitalization. The reform also incites individuals to take more prompt reaction to less
serious diseases, reflected by the effect on outpatient variable. The effect on the bed occupancy ratio
suggests that hospital inputs are better used than that in the period before the NRCMS. The ALOS is
decreased by 8.5% after the introduction of insurance system. To summarize, insurance system
imposes incentives for efficiency on the supplier side. These results are consistent with those
presented in the literature, for example Wagstaff et al (2009).
In terms of financial structure, the share of the business income generated by drug selling
increases by 3.25 percentage points due to NRCMS. We expected a decrease as it is well-known that
over-prescription was in China a real problem. However, the rise of this indicator may be induced by:
a decrease in non drug business income; or ii) a lower increase of that income relatively to drug selling
income. A simple regression of business income (in logarithm) on participation dummy shows that
business income increased before and after the implementation of reform. This implies that the
augmentation of the share of drug income is higher than that of the business income.
4.5 Heterogeneity of the impact
Equation 1 supposed the impact of NRCMS is homogenous. This section investigates the
potential heterogeneous impacts of reform across space and over time. Results are presented in Tables
8A and 8B.
10 Covariates are introduced into the model in a gradual way to test the stability of the coefficient on Pijt. The impact does not
change significantly. Only the regression with all the covariates is shown in table 7. The others are available in request. 11 Participation is a dummy variable so the elasticity is calculated by the following formula: [e(coefficient)-1]*100.
13
First, the incentive brought by the new insurance system is different for the population with
different socio-economic characteristics. It can be expected that the price effect of insurance is less
important for people in rich townships than those in poor townships. In order to check these
differentiated impacts across space, a dummy variable “poor”12
is interact with Pijt. The reform shows
higher impacts on discharged patient in poor townships than in non-poor ones. For the bed occupancy
ratio, the interactive variable is significant and positive whereas the participation dummy is not. It
indicates that the reform has a positive impact on the bed occupancy ratio only for township hospitals
located in poor townships. In contrast, impact is not significantly different for the number of outpatient
visits, the ALOS and the share of drug income.
Second, when a new policy is implemented, it may take time for the target population and local
authority to understand new rules and regulations, to construct trust from the demand side, and to
increase the efficiency of application. It is therefore expected that the insurance has more influence on
local agents’ behaviors in places which have more experience with reform and higher coverage of
insurance. Inspired by Aker (2008), Galiani, Gertler and Schargrodsky (2008), this paper identifies
experience effect in two ways.
First, the participation dummy is replaced by a variable measuring the NRCMS insurance
coverage in township. Coefficients on this variable are significant for all outputs (except for the bed
occupancy rate). If the insurance coverage increases by 10 percentage points, it will increase the
number of outpatient visits by 2.12%, the number of discharged patient by 6.44%, and have almost no
effect on the ALOS (elasticity = 0.94%). Concerning the drug share in total business income, an
increase of 10 percentage points of insurance coverage will increase this outcome by 0.38 percentage
points. The variable representing the coverage of insurance comes out with similar signs and
significant as the participation dummy. So, it conforms to the effects of insurance reform captured by a
single dummy.
Second, the experience is measured by distinguishing the number of years that township is
involved into the reform. For this sake, three dummies are created: “entry year in reform”, “1 year
after the reform”, “2 years or more after the reform”13
. For three outputs out of five, the effects of
reform evolve in the time.
For discharged patients, reform has a positive and significant impact during the year of
implementation as well as the first year after. The effect enforces in the time. Compare to
township hospitals without reform, a rise of 53% of discharged patients is observed in the
12 Poor is calculated from the rural net income per capita of the township. A township is considered as poor when it belongs
to the quintile 1. 13 “entry year in reform” equals to 1 if the township enters into NRCMS in actual period, and 0 otherwise; “1 year after the
reform” equals to 1 if the township has been in the program since one year, 0 otherwise; “2 years or more after the reform”
equals to 1 if reform as implemented in the township since two years or more, 0 otherwise.
14
township hospitals which just enter into reform and 55% for those which have been in the
reform since one year.
Lastly, the share of business income generated by drug selling increased faster during all the
periods after the reform. This suggests that the reform has a significant and persistent effect on
financial structure of township hospitals.
5 Discussion and Conclusion
Chinese government began in 2003 the implementation of the NRCMS, a medical insurance
system for rural areas. This paper analyzes the impact of the introduction of NRCMS in Weifang
prefecture, located in Shandong province. Based on a sample of randomly selected 24 township
hospitals over a nine years period, the impact of NRCMS on township hospitals outcomes is estimated
with fixed effect method, after having checked all the assumptions for the model with staggered entry.
The originality of the paper is to focus the analysis of impact of insurance reform on the service
provider side. The choice of township hospital as the object of this study is relevant for two reasons.
First, township hospitals serve mainly rural residents who are also the target population of NRCMS
reform. This constructs a straight linkage between the insurer (NRCMS) and the service provider
(township hospitals). Second, NRCMS aims at increasing the rural population’s access to basic
medical services, and the township hospitals play an important role in delivering this kind of services
beside village health station (HS), TH are also the first referral facilities above HS and the first one to
deliver general surgical services, before county level. By studying the impact of the implementation of
NRCMS on the activities of township hospital, it permits to measure the distance between political
expectation and practical results and thus gives inspiration for future health and social policy
orientation.
Estimations confirm that insurance reform increases township hospitals activities. First, the
NRCMS has a significant and positive impact on the number of outpatient, the number of discharged
patients, and the bed occupancy ratio and it reduces the ALOS. Second, financial structure also
evolves with the reform, as it can be shown by the increased of the share of the business income
generated by the drug selling but in the worse way.
These results are particularly interesting from a policy point of view. First, they suggest that the
reform incites medical consultation and consequently increase the activities of township hospitals, as
measured by inpatient, outpatient visits and the bed occupancy ratio. Insurance impact is more
important for hospitalization than outpatient visits, because the reimbursement rate is higher for the
former and most important of all because the cost of inpatient care is much higher than for ambulatory
care. This result conforms to the policy aim of using insurance to lower financial obstacle in the access
15
to expensive medical services. Second, results show a significant impact on the ALOS. It decreased
with the introduction of the reform. Insurance can lead hospitals to search for more efficient service
provision by giving correct economic incentive. It is then expected that insurance can encourage
hospitals to reduce unnecessary hospital stay as there is a ceiling on reimbursement and a strong effort
from Health Bureau to prevent increase of prices. The first effect seems to be confirmed by our study.
Concerning the financial structure of township hospitals, it evolves with the reform as we have
seen. Before the reform, the share of drug revenue in business income was the main source of revenue
for township hospitals. Our results highlight that this phenomenon is reinforced by the reform. It is
certainly true that more patients imply more drug prescriptions and so more revenue from this. But the
increased of the revenue from drug selling is more important than that of activities income. But on this
point, a detailed study on overall relevance of drug prescription will be necessary to fully understand
this specific impact of the reform as it could reflect a principal-agent and a moral hazard issues with
some unnecessary drug expenditure.
Finally, estimations show that the impact can be heterogeneous across space and time, especially
for hospitalization. Discharged patients increase more in poor region after the implementation of
insurance than that in non-poor region. Such results are not surprising because income is the principal
obstacle to the hospitalization, and the price elasticity of demand is higher for poor households. The
insurance scheme gives similar financial assistance to all the insured. This represents a more important
proportion of capacity to pay of the poor than non-poor. But we can’t definitely infer from this point
that catastrophic health care expenditures have been drastically reduced, although it could seem
reasonable, as Wagstaff et al (2009) shown that in Gansu medical insurance has lead households to
search for more specialized and costly care, the result being for some of them an increase in
catastrophic expenditure.
Obviously, health care demand to township hospitals has increased thanks to the reform. But we
don’t know to what extend it implies a net increase of demand or if a share of this demand is coming
from a transfer of demand from the county upper level, as NCMS increase the gap between the cost of
care at township and county hospitals.
The influence of the NRCMS also changes over time because of the extension of the coverage
rates and accumulated experience. However a longer period is necessary to assess the long term
impact of the reform.
The main precaution needs to pay attention in this study is the relatively small sample. Only 24
hospitals are involved into the study. Although this shortcoming is partly “compensated” by
randomized process of selection of the hospitals and by the nine years of the survey period, it would
be desirable to check the results with a larger sample. As all the hospitals are coming from Weifang
16
prefecture, the results cannot be directly generalized for other regions in China, especially in poor
provinces as Ningxia or Qinghai for example.
References
Aker Jenny C., (2008), “Does digital divide or provide? The impact of cell phones on grain markets in
Niger”, Center for Global Development Working Paper No. 154.
Barham T., (2005), “Providing a healthier start to life: The impact of conditional cash transfers on
infant mortality”, Working paper No. 06-02.
De Janvry A., McIntosh C. and Sadoulet E., (2009), “The Supply- and Demand-Side Impacts of Credit
Market Information”, Journal of Development Economics, forthcoming.
Eggleston K., Ling L., Qingyue M., Lindelow M. and Wagstaff A., (2008), “Health Service delivery in
China: a literature review”, Health Economics, 17, pp. 149-165.
Galiani S., Gertler P., and Schargrodsky E., (2008), “School decentralization: helping the good get
better, but leaving the poor behind”, Journal of Public Economics, 92(10-11), pp. 2106-2120.
Gruber J. and Hanratty M., (1995), “The Labor-Market Effects of Introduction National Health
Insurance: Evidence from Canada”, Journal of Business & Economic Statistics, 13(2), pp. 163-173.
Hillier S. and Shen J., (1996), “Health care systems in transition: People's Republic of China. Part I:
An overview of China's health care system”, Journal of Public Medicines, 18(3), pp. 258-265.
Hsiao W., (1984), “Transformation of health care in China”, The New England Journal of Medicine,
310(14), pp. 932-936.
Hsiao W., (1995), “The Chinese Health Care system: lessons for other nations.” Social Sciences &
Medicine, 41(8), pp. 1047-1055.
Imbens, G., and Wooldridge J., (2007), “What’s New in Econometrics.” Lecture Notes, NBER
Summer Institute.
Liu Y., Rao K. and Hu S., (2003), “People’s Republic of China. Toward Establishing A Rural Health
Protection System”, Beijing: Asian Development Bank.
Liu J., Xu L., Cao X., Wang X., Sun H., Tang C., Yang Z., Song X., Lv M., (2008), “Analysis of
satisfaction about new cooperative medical scheme and its influencing factors in Weihai, China”,
Health Policy, 86, pp. 239-244.
Liu, X., L. Xu, and Wang S., (1996), “Reforming China’s 50,000 Township Hospitals: Effectiveness,
Challenges, and Opportunities.” Health Policy, 38 (1), pp. 13-29.
Mao Z., (2006), “Pilot program of China’s new cooperative medical scheme system design and
progress.” United Nations for Economic and Social Commission for Asian and the Pacific,
http://www.unescap. org/esid/hds/issues/index.asp.
18
Matteo L. D., (2003), “The Income Elasticity of Health Care Spending: A Comparison of Parametric
and Nonparametric Approaches.” , The European Journal of Health Economics, 4(1), pp. 20-29.
Mocan H. N. and Tekin E., (2004), “The Demand for Medical Care in Urban China”, World
Development, 32(2), pp. 28-304.
Sepehri A., Sarma S., Serieux J., (2009), “Who is giving up the free lunch? The insured patients'
decision to access health insurance benefits and its determinants: Evidence from a low-income
country”, Health policy, 92(2-3), October, pp.250-258.
Wagstaff A., Lindelow M., Jun G., Ling X., and Juncheng Q., (2009), “Extending health insurance to
the rural population: An impact evaluation of China’s new cooperative medical scheme”, Journal of
Health Economics, 28, pp. 1-19.
Wagstaff A. and Moreno-Serra R., (2009), “Europe and central Asia’s great post-communist social
health insurance experiment: Aggregate impacts on health sector outcomes”, Journal of Health
Economics, 28, pp. 322-340.
Wooldridge JM., (2002), Econometric analysis of cross section and panel data. Cambridge, MIT
Press.
World Bank. 1997. Financing Health Care: issues and options for China. China 2020 series.
Washington D.C.: World Bank.
World Bank. 2009. Reforming China’s Rural Health System. Direction in development Human
development. Washington D.C.: World Bank.
Xueshang F., Shenglan T., Bloom G., Segall M. and XingYuan G., (1995), “Cooperative medical
schemes in contemporary rural China”, Social science & Medicine, 41(8), pp. 1111-1118.
Yip W., and Hsiao William C., (2008), “The Chinese health system at a crossroads”, Health Affairs,
27(2), pp.460-468.
Yip W., and Hsiao William C., (2009), “China's health care reform: A tentative assessment”, China
Economic Review, 20, pp. 613-619.
Yip W., and Hsiao William C., (2009), “Non-evidence-based policy: How effective is China’s new
cooperative medical scheme in reducing medical impoverishment?”, Social Science & Medicine, 68,
pp.201-209.
You X., Kobayashi Y., (2009), “The new cooperative medical scheme in China”, Health Policy, 91,
pp. 1-9.
19
Table 1: NRCMS participation from 2004 to 2008
Year Number of counties
which participate
Number of persons which are
enrolled (100 million persons) Rate of participation
2004 333 0.8 75.2
2005 678 1.79 75.66
2006 1451 4.1 80.66
2007 2451 7.26 86.2
2008 2729 8.15 91.53
Source: China data online, China Statistical Yearbook 2009.
Table 2: Phasing-in of the rural health insurance reform in the studied townships
County Township 2003 2004 2005 2006
Anqui (2003)
Jin Zhongzi
Guan Gong Zhen
Jingshi
Lin Wu
Wu Shan Zhen
Xin An
Zhe Shan Zhen
Changyi
(2004)
Liu Tan
Xia Dian Zhen
Gaomi (2005)
Cai Gou Zhen
Da Mou Jia Zhen
Jing Gou Zhen
Kan Jia Zhen
Qinzhou
(2003)
Dong Xia Zhen
Gao Liu Zhen
He Guan Zhen
Shao Zhuang Zhen
Tan Fang Zhen
Wang Fen Zhen
Shouguang
(2004)
Dao Tian Zhen
Hou Zhen
Tian Liu Zhen
Zhucheng
(2004)
Bai Chi He
Ma Zhuang Zhen Note: the entry date of county is in the parentheses. In our sample, not all townships entered at the same date
that the county which they belong.
The year that the township is covered by NRCMS.
Source: Authors’ database.
20
Table 3a: Descriptive statistics I
Total sample (2000 – 2008) YEAR 2000 YEAR 2003 YEAR 2008
Variables Obs Mean Std. Dev. Min Max Obs Mean Std. Dev. Min Max Obs Mean Std. Dev. Min Max Obs Mean Std. Dev. Min Max
Township characteristics’
# of households14 216 1.36 0.68 0.59 3.98 24 1.09 0.54 0.59 3.33 24 1.26 0.61 0.63 3.34 24 1.79 0.83 0.70 3.98
Share of rural (%) 216 92.80 4.38 78.62 98.65 24 93.46 3.98 82.00 98.65 24 93.56 4.17 83.00 98.65 24 92.57 4.46 79.62 98.04
Farmer net income
per capita15 211 0.45 0.12 0.07 0.83 22 0.33 0.05 0.20 0.41 23 0.38 0.06 0.23 0.47 24 0.63 0.09 0.45 0.83
# of health stations 216 31.22 17.87 8 97 24 25.33 15.77 8 71 24 29.50 16.68 8 71 24 37.25 19.70 8 75
Township hospitals characteristics’
# of beds 213 40.62 19.17 16 150 23 34.78 13.02 16 65 24 35.92 13.43 16 70 24 59.08 32.35 20 150
# of professionals 213 51.64 31.49 16 201 23 46.00 25.83 18 143 24 49.13 29.61 17 159 24 66.96 42.32 21 201
- # of seniors 205 1.15 1.29 0 7 22 1.09 1.19 0 4 23 1.17 1.53 0 6 23 1.17 1.11 0 4
- # of intermediate 213 8.30 8.19 0 42 23 5.96 7.16 0 37 24 7.25 7.36 0 39 24 12.79 9.90 0 37
- # of juniors 213 41.16 23.60 13 171 23 37.70 19.12 15 102 24 39.67 22.33 14 114 24 52.17 33.97 18 171
# of outpatient
visits 213 30196.21 24882.01 4934.00 138911 23 25395.26 21332.54 5432 98208 24 25350.04 21182.67 5508 78172 24 48244.58 32582.50 7082 138911
# of discharged
patients 213 1271.42 892.17 54.00 4596 23 958.87 642.79 56 3369 24 912.29 712.21 94 3422 24 2184.00 1077.56 738 4467
# of actual
vaccination 213 9122.75 9523.00 103.00 73378 23 6795.26 4991.84 156 18507 24 7362.79 5329.86 103 20958 24 14860.88 18529.26 108 73378
ALOS (in days) 216 5.03 1.42 2.14 15.2 24 5.29 2.43 3 15.2 24 5.21 1.52 2.14 9.96 24 5.12 0.92 3.2 7
Bed occupancy
ratio (in days) 216 0.47 0.24 0.03 1.40 24 0.44 1.25 0.03 1.25 24 0.38 0.24 0.05 0.91 24 0.61 0.19 0.26 0.94
14
Unit: 10000 households. 15
All monetary variables are in 10000 Yuan and normalized to 2000 constant prices.
21
Table 3b: Descriptive statistics II
Total sample YEAR 2000 YEAR 2003 YEAR 2008
Variables16 Obs Mean Std. Dev. Min Max Obs Mean Std. Dev. Min Max Obs Mean Std. Dev. Min Max Obs Mean Std. Dev. Min Max
Total subsidies 212 16.55 22.07 0 168.28 23 15.54 14.78 0 64.64 24 14.26 14.62 0.00 49.41 24 37.17 47.60 1.73 168.28
County subsidies 211 11.11 19.34 0 168.28 23 9.45 9.67 0 34.56 24 9.65 12.11 0.00 43.48 24 27.48 47.43 0 168.28
Government
subsidies 216 2.61 12.55 0 103.87 24 1.70 8.32 0 40.74 24 2.06 10.09 0.00 49.41 24 5.95 22.30 0 103.87
Business income 216 202.48 195.40 18.65 1265.54 24 133.28 99.72 34.38 486.78 24 129.34 99.81 33.37 380.54 24 418.69 311.31 48.42 1265.54
Medical treatment
income 216 78.99 88.29 5.47 560.65 24 50.42 47.24 12.55 207.96 24 48.40 47.31 5.47 182.17 24 152.80 138.06 19.61 560.65
Drug income 216 123.49 112.18 12.52 704.89 24 82.85 56.77 21.83 278.82 24 80.94 56.28 27.90 226.18 24 265.88 181.39 28.81 704.89
Total expenditures 216 290.92 251.65 41.96 1635.84 24 191.00 112.74 86.97 599.35 24 216.12 143.13 69.65 636.77 24 569.56 396.79 143.93 1635.84
Current
expenditures 207 284.62 246.71 41.96 1583.82 23 182.75 111.75 83.67 573.13 23 213.85 137.44 69.65 603.66 23 553.44 396.28 129.51 1583.82
Staff expenditures 215 112.24 93.78 19.08 639.8 24 66.96 35.70 28.72 179.89 24 87.48 55.19 32.07 244.27 24 204.42 142.48 30.75 639.80
Drug expenditures 215 78.32 71.78 0 528.05 24 65.70 49.61 0 235.28 24 50.74 38.59 11.42 189.24 24 161.98 116.04 36.70 528.05
NRCMS
reimbursement 127 86.22 114.74 0 699.96 4 0 0 0 0 10 26.93 32.03 0 91.29 24 190.96 182.17 36.47 699.96
for inpatient 127 32.84 48.91 0 226.63 4 0 0 0 0 10 15.91 20.24 0 51.68 24 69.22 71.50 4.29 226.63
for outpatient 127 53.11 71.57 0 473.56 4 0 0 0 0 10 11.32 15.05 0 39.61 24 120.18 119.63 18.56 473.56
16
All monetary variables are in 10000 Yuan and normalized to 2000 constant prices.
22
Table 4: Test of parallel trend hypothesis
Explained Variables
Pre-intervention years Ln (outpatient) Ln (discharged patient) Ln (length) Bed occupancy rate Share of drug income
Township hospitals which implemented NRCMS in 2003
2001 -0.0532 0.140 0.0333 -0.00308 -0.346
(0.276) (0.306) (0.113) (0.111) (3.583)
2002 -0.129 0.119 0.0208 0.0383 2.746
(0.250) (0.277) (0.123) (0.137) (4.906)
Township hospitals which implemented NRCMS in 2004
2001 0.273 -0.302 -0.144 0.223 -1.556
(0.277) (0.581) (0.129) (0.194) (5.765)
2002 0.400 -0.527 -0.0545 0.122 -6.093
(0.240) (0.617) (0.134) (0.237) (6.580)
Township hospitals which implemented NRCMS in 2005
2001 0.123 -0.377 0.0942 -0.0972 -1.622
(0.480) (0.519) (0.0938) (0.130) (4.309)
2002 0.140 -0.427 0.0773 -0.137 -7.065
(0.438) (0.577) (0.105) (0.153) (4.823)
Township hospitals which implemented NRCMS in 2006
2001 0.00754 0.223 -0.0130 -0.0723 3.838
(0.384) (0.372) (0.127) (0.115) (3.802)
2002 0.000256 0.0139 -0.0636 -0.131 3.140
(0.370) (0.388) (0.171) (0.146) (5.469)
Note: Standard errors (in parentheses) are corrected for heteroskedasticity. *** indicates significance at 1%; ** at 5%; and, * at 10%.
Test is performed on a sample from 2000 to 2003.
23
Table 5: Test of exogeneity of the order of NRCMS implementation
Explained Variables
Ln (outpatient) Ln (discharged patient) Ln (length) Bed occupancy rate Share of drug income
1 year prior to NRCMS 0.0280 0.132 -0.0382 0.0315 -0.365
(0.400) (0.655) (0.201) (0.126) (3.499)
2 years prior to NRCMS 0.0784 0.108 -0.0101 0.0155 -1.674
(0.287) (0.429) (0.126) (0.0970) (3.150)
Observations 72 72 72 72 72 Note: Standard errors (in parentheses) are corrected for heteroskedasticity. *** indicates significance at 1%; ** at 5%; and, * at 10%.
The five specifications are estimated on a sample consisted of the three years preceding the entrance date into the reform of each township. They also included year and
county fixed effects.
24
Table 6: Comparison of means between phasing-groups for pre-reform year (i.e. 2002)
Year reform 2003 Year reform 2004 Mean
2003 -
Mean
2004
Diff/sd.
(2003)
Year reform 2005 Mean
2003 -
Mean
2004
Diff/sd.
(2003)
Year reform 2006 Mean
2003 -
Mean
2004
Diff/sd.
(2003)
Obs. Mean
Std.
Dev. Obs. Mean
Std.
Dev. Obs. Mean
Std.
Dev. Obs. Mean
Std.
Dev.
Township characteristics
Surface 6 110.453 61.510 5 113.040 63.168 -2.587 -0.042 5 99.114 17.488 11.339 0.184 8 99.083 45.688 11.371 0.185
# of Households 6 1.075 0.261 5 1.496 0.712 -0.421 -1.615 5 1.160 0.330 -0.085 -0.326 8 1.279 0.861 -0.204 -0.781
Share of rural population 6 93.552 3.457 5 94.040 2.876 -0.488 -0.141 5 93.632 5.537 -0.080 -0.023 8 93.354 4.902 0.198 0.057
Rural net income per capita 6 0.317 0.063 5 0.372 0.084 -0.055 -0.872 5 0.342 0.019 -0.026 -0.405 8 0.350 0.080 -0.033 -0.521
Township hospital characteristics
# of operational beds 6 30.167 5.307 5 34.600 12.896 -4.433 -0.835 5 35.000 12.247 -4.833 -0.911 8 34.750 16.175 -4.583 -0.864
# of medical professional 6 26.833 7.985 5 36.800 16.814 -9.967 -1.248 5 39.000 22.572 -12.167 -1.524 8 36.625 39.038 -9.792 -1.226
Government subsidies17
6 0.000 0.000 5 8.902 19.906 -8.902 NC 5 0.000 0.000 0.000 NC 8 0.000 0.000 0.000 NC
County subsidies 6 2.390 3.316 5 5.856 9.116 -3.465 -1.045 5 13.854 3.010 -11.463 -3.457 8 15.455 13.548 -13.065 -3.939
17
Diff/sd.(2003) cannot be calculated because standard deviation equals zero. But group which enters in the reform in 2004 is largely different than the 3others, so we consider government
subsidies as a covariate.
25
Table 7: Impact of the NRCMS
Variables in Logarithm Variables in level
Outpatient Discharged
patient ALOS
Bed occupancy
ratio
Share of drug
income
Participation 0.160* 0.504*** -0.0818* 0.0753* 3.251*
(0.095) (0.18) (0.044) (0.0407) (1.76)
# of households 0.123 0.285 -0.0609 -0.0546 0.282
(0.17) (0.24) (0.044) (0.0635) (1.98)
Rural net income per capita -1.596 3.677** -0.450 0.583 -1.804
(0.97) (1.56) (0.66) (0.507) (17.9)
# of medical professionals 0.000253 0.00179 -0.000806 -0.000487 -0.0312
(0.0034) (0.0066) (0.0018) (0.00212) (0.070)
# of operational beds 0.000822 0.00601 0.00402*** -0.00197 -0.109*
(0.0044) (0.0062) (0.0013) (0.00191) (0.064)
Government subsidies -0.0103*** 0.0164 -0.00585** -0.00619** -0.155***
(0.0032) (0.018) (0.0027) (0.00301) (0.055)
County subsidies -0.00319* -0.00657** -0.00203* -0.000758 0.0897**
(0.0019) (0.0032) (0.0011) (0.000952) (0.039)
Constant 9.827*** 4.312*** 2.162*** 0.358** 76.36***
(0.37) (0.59) (0.25) (0.176) (5.60)
Observations 215 215 215 215 215
Note: Standard errors (in parentheses) are corrected for heteroskedasticity and AR(1) serial correlation. ***
indicates significance at 1%; ** at 5%; and, * at 10%.
26
Table 8A: Heterogeneous impact across time of the NRCMS for dependant variables in logarithm
Variables in logarithm
Outpatient Discharged Patient ALOS
Participation 0.122 0.401** -0.0748*
(0.095) (0.18) (0.042)
NRCMS coverage 0.00212* 0.00644*** -0.0009427*
(0.0012) (0.0024) (0.00053)
Entry year to reform 0.1257 0.4251** -0.0628
(0.0885) (0.1881) (0.0449)
1 year after reform 0.1317 0.4369** -0.0887
(0.1219) (0.2087) (0.0577)
2 years or more after reform -0.0049 0.2337 -0.0435
(0.151) (0.2398) (0.0782)
Participation*Poor 0.169 0.460*** -0.0315
(0.11) (0.15)
(0.064)
Note: Standard errors (in parentheses) are corrected for heteroskedasticity and AR(1) serial correlation. *** indicates significance at 1%; ** at 5%; and, * at 10%.
27
Table 8B: Heterogeneous impact across time of the NRCMS for dependant variables in level
Variables in level
Bed occupancy rate Share of drug income
Participation 0.0457 3.751**
(0.0404) (1.83)
NRCMS coverage 0.000762 0.0378*
(0.000515) (0.023)
Entry year to reform 0.0351 3.1403**
(0.0398) (1.5952)
1 year after reform 0.0661 4.9578**
(0.0525) (2.5083)
2 years or more after reform -0.0572 5.5840*
(0.0597) (3.0039)
Participation*Poor 0.1329*** -2.244
(0.0459) (2.34)
Note: Standard errors (in parentheses) are corrected for heteroskedasticity and AR(1) serial correlation. *** indicates significance at 1%; ** at 5%; and, * at 10%.