Economies of scale and scope in Vietnamese hospitals
-
Upload
marcia-weaver -
Category
Documents
-
view
221 -
download
0
Transcript of Economies of scale and scope in Vietnamese hospitals
Social Science & Medicine 59 (2004) 199–208
ARTICLE IN PRESS
*Correspond
206-221-4945.
E-mail addr
0277-9536/$ - se
doi:10.1016/j.so
Economies of scale and scope in Vietnamese hospitals
Marcia Weavera,*, Anil Deolalikarb
aDepartment of Health Services, University of Washington, 901 Boren Avenue, Suite 1100, Seattle, WA 98104, USAbDepartment of Economics, Sproul Hall, University of California, Riverside, CA 92521, USA
Abstract
Hospitals consume a large share of health resources in developing countries, but little is known about the efficiency of
their scale and scope. The Ministry of Health of Vietnam and World Bank collected data in 1996 from the largest
sample ever surveyed in a developing country. The sample included 654 out of 815 public hospitals, six categories of
hospitals and a broad range of sizes.
These data were used to estimate total variable cost as a function of multiple products, such as admissions and
outpatient visits. We report results for two specifications: (1) estimates with a single variable for beds and (2) estimates
with interaction terms for beds and the category of hospital. The coefficient estimates were used to calculate marginal
costs, short-run returns to the variable factor, economies of scale, and economies of scope for each category of hospital.
There were important differences across categories of hospitals. The measure of economies of scale was 1.09 for
central general and 1.05 for central specialty hospitals with a mean of 516 and 226 beds, respectively, indicating roughly
constant returns to scale. The measure was well below one for both provincial general and specialty hospitals with a
mean of 357 and 192 beds, respectively, indicating large diseconomies of scale. The measure was 1.16 for district
hospitals and 0.89 other ministry hospitals indicating modest economies and diseconomies of scale, respectively. There
were large economies of scope for central and provincial general hospitals.
We conclude that in a system of public hospitals in a developing country that followed an administrative structure,
the variable cost function differed significantly across categories of hospitals. Economies of scale and scope depended
on the category of the hospital in addition to the number of beds and volume of output.
r 2003 Elsevier Ltd. All rights reserved.
Keywords: Economies of scale; Economies of scope; Hospital cost functions; Vietnam
Introduction
Hospitals consume the largest share of health
resources in most countries. Barnum and Kutzin
(1993) reported that hospitals received 50 percent or
more of government health resources in 19 out of 29
developing countries for which data were available, and
hospitals received on average 54 percent of government
health resources in OECD countries. Hospitals received
54.3 percent of government health funds in Vietnam in
1996 (Le Duc Chinh, undated).
ing author. Tel.: +1-206-616-9173; fax: +1-
ess: [email protected] (M. Weaver).
e front matter r 2003 Elsevier Ltd. All rights reserve
cscimed.2003.10.014
Many public health experts believe that some of those
resources would be better spent on preventive and
primary care (World Bank, 1993). Even within the
hospitals’ share, questions are raised about the efficiency
of their scale and scope. The question about scale is
whether larger hospitals are more or less efficient than
smaller ones. On the one hand, hospitals require large
investments in capital such as buildings, equipment and
specialized staff, which may make it more efficient to
have one large hospital rather than two small ones. On
the other hand, hospitals are complex organizations to
manage, and at some point a smaller hospital may run
more smoothly than a larger one.
The question about scope is whether or not it is
efficient to combine outpatient and inpatient care at the
d.
ARTICLE IN PRESSM. Weaver, A. Deolalikar / Social Science & Medicine 59 (2004) 199–208200
same facility. Physicians often need to see patients on
both an inpatient and an outpatient basis; an outpatient
who receives diagnostic exams may later be admitted or
an inpatient that is discharged may need follow-up
visits. In some cases, it may be more efficient for
physicians to provide both types of care from a single
office at the hospital. In other cases, it may be more
efficient to reduce the daily flow of a large volume of
outpatients at the hospital by having separate facilities.
These questions may be answered with estimates of a
hospital cost function that shows the relationship
between cost and output. Although researchers have
estimated cost functions for hospitals in developing
countries (Anderson, 1980; Barnum & Kutzin, 1993;
Bitran-Dicowsky & Dunlop, 1993) and for non-hospital
health facilities in Nigeria (Wouters, 1993), their
research was limited by samples of a relatively small
number of facilities. The sample sizes ranged from eight
to 51 facilities. Even in pooled cross-section and time-
series samples, the largest sample size was 72 observa-
tions (Barnum & Kutzin, 1993). Barnum and Kutzin
(1993) recommended further research to examine
differences in hospital cost functions by hospital level
with a large sample of hospitals.
The Ministry of Health of Vietnam and the World
Bank conducted a survey of hospitals to collect
information on hospital activity, revenue and costs.
The survey provided 1996 data on one of the largest
samples of hospitals ever collected in a developing
country, with information on 654 out of 815 public
hospitals in Vietnam. The sample included six categories
of hospitals: four levels of hospitals (central, provincial,
district and other ministry) and two classes of hospitals
(general and specialty) at the central and provincial
level, as well as a broad range of sizes of hospitals as
measured by number of beds, admissions and out-
patients.
After a brief background section on hospital reforms
in Vietnam, we report estimates of the hospital variable
cost function using the data from the survey of hospitals
in Vietnam. These estimates were used to calculate
marginal costs, short-run returns to the variable factor
(SRVF), economies of scale, and economies of scope for
the six categories of hospitals.
Hospital reforms in Vietnam
The survey of hospitals occurred during a period of
rapid transformation of the health sector in Vietnam.
Historically, health care in Vietnam was provided
exclusively by public facilities funded by the government
or commune health centers funded by work brigades.
Hospitals were allocated by administrative units, with
central level hospitals in major cities, provincial
hospitals in provincial capitals, and district hospitals in
district capitals. The government specified the services
that would typically be offered at each level.
Three health sector reforms in Vietnam in the late
1980s and early 1990s affected hospitals: (1) the
introduction of user fees in hospitals, (2) the introduc-
tion of social insurance in 1993, and (3) a 1993
ordinance that legalized private medical practice and
pharmacies (World Bank, 2001). By 1996, revenue from
user fees and social insurance accounted for 35 percent
of hospital income, with the shares ranging from 39
percent of income at central general hospitals to 25 and
27 percent at central specialty and district hospitals,
respectively (World Bank, 2001; Weaver & Rubanowice,
1999). At the same time, real growth rates in government
spending on hospitals were low or negative. By 1996,
patient revenue was almost as large as a percentage of
income as government spending for central and provin-
cial general hospitals.
The growth in patient revenue affected hospital
expenditures on staff and drugs (World Bank, 2001;
Weaver & Rubanowice, 1999). Hospitals were allowed
to spend up to 35 percent of patient revenue on staff
bonuses and had discretion over how to allocate them.
By 1996, staff bonuses accounted for 30 percent of
personnel expenditures at hospitals, with shares ranging
from 47 percent at central general and provincial
specialty hospitals to 12 percent at district hospitals.
At the same time, the real growth rates of salaries were
negative. By 1996, bonuses were a larger percentage of
personnel expenditures than salaries at provincial
specialty hospitals. Expenditures on drugs also increased
and in 1996 accounted for 34 percent of expenditures at
hospitals. By 1996, expenditures on drugs exceeded
expenditures on personnel at central and provincial
hospitals.
The legalization of private medical practice did not
increase the number of private hospitals. In 1996, there
were only four private hospitals in Vietnam and this was
still true as recently as 2001 (World Bank, 2001). It did
increase the number of health personnel with licenses to
work in private practice. In 1996, 26,000 health workers
or about one-tenth of the health work force had licenses
for private practice. Almost half of them were public
employees who worked in the private sector during
evenings and weekends (World Bank, 2001).
Methods
Model
The hospital cost function estimates were based on a
multiple-product model, in which total costs were a
function of input prices and the level of output of
multiple products, such as inpatient days and outpatient
visits (Grannemann, Brown, & Pauly, 1986). Two
ARTICLE IN PRESSM. Weaver, A. Deolalikar / Social Science & Medicine 59 (2004) 199–208 201
features of the multiple-product model were: (1)
coefficients were estimated for each product, as opposed
to having to create a single index of products (e.g.
converting four outpatient visits into one inpatient day),
and (2) the model included interaction terms for
products to measure the economies of scope.
We used an extension of that model in which total
variable costs were a function of the number of hospital
beds as a proxy measure of the capital stock, as well as
input prices and multiple outputs (Vita, 1990). The
model was based on the assumption that hospitals could
use variable inputs like personnel and medical supplies
at cost-minimizing levels, but could not adjust the
number of beds in the short-run. Aletras (1999)
explained that the coefficient for the beds variable offers
a test of that assumption; a coefficient that was positive
and significant implied that the hospitals were not at
their long run equilibrium. Aletras (1999) tested the
assumption using data on public hospitals in Greece,
and concluded that the total variable cost model was
preferable to the total cost model.
The equation for the total variable cost function was
C ¼ ea0þa1bedsef ðY ;X Þ; ð1Þ
where Y is a vector of hospital products, such as
admissions, outpatient visits, and diagnostic exams. X is
a vector of independent variables that shift the cost
function, such as the region in which the hospital was
located and the category of the hospital. Input prices
were not included for reasons that are explained in the
variables subsection below.
The equation was estimated by taking the natural log
of both sides. It had a flexible functional form with
linear, squared and cubed values for admissions and
outpatients and an interaction term between admissions
and outpatient visits:
ln C ¼ a0 þ a1bedsþ b11Y1 þ b21Y 21 þ b31Y 3
1 þ b12Y2
þ b22Y 22 þ b32Y 3
2 þ b1�2Y1Y2
þXn
k¼3
bkYk þXm
l¼1
clXl ; ð2Þ
where Y1 is the number of admissions and Y2 is the
number of outpatient visits.
The coefficient estimates of Eq. (2) were used to
calculate marginal costs and measures of economies of
scale and scope. The marginal cost of admissions or
outpatient visits (or any product with linear, squared,
cubed and interaction variables) was
MCi ¼ Cðb1i þ 2b2iYi þ 3b3iY2i þ b1�2YjÞ; ð3Þ
where j is the outpatient visits if i is the admissions and
vice versa. The marginal cost of operations and
diagnostic exams (or any product with a linear variable
and no interaction) was
MCk ¼ Cbk: ð4Þ
Following Vita (1990) and Barnum and Kutzin
(1993), we derived measures of economies of scale and
scope. Eqs. (5a)-(7a) below are from Barnum and
Kutzin (1993). We substituted the term for total variable
cost from Eq. (2) and for marginal costs from Eqs. (3)
and (4) above into those equations. Then, we used
algebra to simplify the equations and obtain the
measures presented in Eqs. (5b)-(7b).
There were short-run and long-run measures of the
relationship between cost and scale. In the short run, the
number of beds was unchanged, and any gains in
efficiency from increasing output accrued to the variable
factors. These gains were called SRVF and measure how
cost changes as output increases when output mix and
the number of hospital beds were unchanged. The
equation for SRVF in Barnum and Kutzin (1993) was
SRVF ¼CPm
i¼1 MCi Yi
� �: ð5aÞ
With our functional form, the measure of SRVF was
SRVF
¼1
P2i¼1 Yiðb1i þ 2b2iYi þ 3b3iY
2i þ b1�2YjÞ þ
PmK¼3 Ykbk
h i:
ð5bÞ
If SRVF was greater than one, the level of output was
less than the most efficient level. If it was less than one,
the level of output was greater than the most efficient
level.
In the long run, the number of beds can change as well
as other factors to improve efficiency. Gains in efficiency
were called economies of scale (EOS) and the equation
for it in Barnum and Kutzin (1993) was
EOS ¼ð1� sC;bedsÞPm
i¼1 sC;Yi
� �;
ð6aÞ
where sa;b is the elasticity of a with respect to b: For thetotal variable cost function, sC;Yi
is the product of the
marginal cost of Yi and the ratio Yi=C: With our
functional form the measure of EOS was
EOS
¼ð1� a1bedsÞP2
i¼1 Yiðb1i þ 2b2iYi þ 3b3iY2i þ b1�2YjÞ þ
PmK¼3 Ykbk
h i;
ð6bÞ
and interpreted the same way as SRVF.
Economies of scope measured the relationship be-
tween cost and product mix. In the hospital cost
function, it measured whether it was less expensive to
provide both inpatient and outpatient care at the
ARTICLE IN PRESSM. Weaver, A. Deolalikar / Social Science & Medicine 59 (2004) 199–208202
hospital than to have a separate facility for outpatients.
The equation for economies of scope in Barnum and
Kutzin (1993) was
Scopes ¼½CðYsÞ þ CðYn�sÞ � CðY Þ�
CðY Þ: ð7aÞ
With our functional form, the measure of economies
of scope was
Scope2 ¼�2b1�2Y1Y2P2
i¼1 Yiðb1i þ 2b2iYi þ 3b3iY2i þ b1�2YjÞ
: ð7bÞ
If scope was greater than zero, it was more efficient to
jointly provide inpatient and outpatient care. If it was
less than zero, it was more efficient to provide them
separately. As shown, the sign of this measure was
determined by the coefficient b1�2; a negative coefficient
indicated economies of scope and a positive coefficient
indicated diseconomies.
Variables
The dependent variable, total variable cost, was total
expenditures on staff, drugs and medical supplies,
maintenance and repairs, and other expenses. The three
types of staff expenditures were salaries, allowances, and
bonuses. All of the numbers included both permanent
and contracted staff. Other expenses included sanitation,
utilities, gasoline, medical records, continuing education
and business travel. Other expenses may have included
some investment to the extent that continuing education
was an investment in human capital. Equipment
purchases were not included.
Table 1
Descriptive statistics by category of hospital
Total Central
general
Cent
spec
Sample means
Cost in VND (millions) 2540 19,900 5560
Cost in $USa 230,657 1,807,120 504,
Beds (hundreds) 1.39 5.61 2.26
Average length of stay 11.86 15.00 50.5
Occupancy rate% 94% 102% 100%
Admissions (thousands) 6.48 16.04 4.34
Outpatients (thousands) 44.00 119.34 29.2
Admissions�outpatients 556.07 2403.87 196.
Operations (thousands) 1.27 3.98 2.78
X-rays (thousands) 8.63 103.02 11.6
Lab tests (thousands) 64.38 566.13 81.6
Sample total
Admissions (thousands) 3868 160 78
Admissions as percentage of total
sample
4% 2%
Sample size 597 10 18
aThe exchange rate on July 1, 1996 was 11,012 Vietnamese Dong
As mentioned in the Model subsection, a variable for
beds was included as a measure of the capital stock.
Beds were measured in units of 100. We report results
for two specifications: (1) estimates with a single variable
for beds and (2) estimates with interaction terms for
beds and the category of hospital that allowed the
coefficient for the beds variable to vary across categories
of hospitals.
We explored two measures of inpatient care: admis-
sions and inpatient days. We selected the admissions
variable, because it was more appropriate for hospitals
with high occupancy rates. As shown in the sample
statistics in Table 1, the average occupancy rate was 94
percent and ranged from 106 percent at other ministry
hospitals to 90 percent at district hospitals. Under these
conditions, more admissions may reflect greater effi-
ciency, whereas inpatient days may be more or less
determined by the number of beds. In fact, the Pearson
correlation between beds and inpatient days was 0.98
compared to a correlation between beds and admissions
of 0.82.
Both measures of inpatient care were potentially
collinear with the beds variable, but the estimates with
the admissions variable performed better. The condition
index was 53.94 for estimates with admissions in Table 4
and 73.07 for estimates with inpatient days. The pattern
of results across categories of hospitals however, did not
differ substantially between the estimates with admis-
sions and those with inpatient days.
To account for differences in case mix across
hospitals, we included the index of case complexity
developed by Roemer, Moustafa, and Hopkins (1968).
ral
ialty
Provincial
general
Provincial
specialty
District Other
ministry
6820 4040 916 1410
904 619,324 366,873 83,182 128,042
3.57 1.92 0.76 0.84
9.06 35.9 6.17 8.96
102% 96% 90% 106%
17.52 6.71 4.36 3.58
3 96.15 49.22 34.71 7.05
79 2053.12 1066.28 178.24 24.63
4.47 2.16 0.38 0.62
3 27.58 13.24 1.90 3.04
8 198.98 99.00 20.83 24.43
1331 496 1678 121
34% 13% 43% 3%
76 74 385 34
(VND) per US dollar ($US) (Jei Corporation, 2002).
ARTICLE IN PRESSM. Weaver, A. Deolalikar / Social Science & Medicine 59 (2004) 199–208 203
The index was
indexi ¼ ALOSiðOCCi=OCCsÞ; ð8Þ
where ALOSi was the average length of stay in hospital
i; OCCi was the occupancy rate of hospital i and OCCs
was the mean occupancy rate for the sample hospitals.
The index used ALOS as a measure of case complexity
and adjusted for exogenous supply and demand
influences that also affected length of stay with the
ratio of OCCi=OCCs:We explored estimates that distinguished admissions
to general beds from admissions to intensive and critical
care unit beds. None of the coefficients for intensive and
critical care unit admissions were significant, so the
results for total admissions were reported below.
Other output variables were outpatient visits, opera-
tions, lab tests, and X-rays. In estimates with higher
level terms for outpatient visits, the coefficients of the
outpatient-squared and outpatients-cubed variables
were not significant and a likelihood ratio test
(LR ¼ 1:32; p-value ¼ 0:52) rejected the specification
with those variables. Estimates with only the outpatient
visits variable were reported below. All of the output
variables were measured in units of 1000.
Finally, the vector of independent variables that
shifted the cost function included variables for each
category of hospital with the exception of provincial
general hospitals and for each region of the country with
the exception of the northern mountains.
The model did not include the prices of inputs, as
originally suggested by Grannemann et al. (1986). Input
price data were rarely available for estimates of hospital
cost functions. For public hospitals in Vietnam, salaries
were established by the government. To the extent that
government salaries were the same across hospitals, the
omission of input prices from the estimates did not bias
the other results. Staff bonuses however, varied across
hospitals. As explained in the section on Hospital
Reforms in Vietnam, hospitals were allowed to spend
up to 35 percent of patient revenue on staff bonuses and
had discretion over how to allocate them. We explored
whether or not staff bonuses could serve as a measure of
input prices in estimates that included the natural log of
the percentage of staff expenditures from bonuses. The
coefficient for that variable was not significant. The
results are not reported below, because the variable
for staff bonuses had a large number of missing values
that reduced the sample size or would have required
recoding.
Similarly, Grannemann et al. (1986) included sources
of revenue as independent variables that shifted the cost
function. It was possible that Vietnam’s new social
insurance program affected hospital costs. We explored
that possibility in estimates that included the percentage
of revenue from social insurance. The coefficient for that
variable was not significant. Those results were not
reported below, because this variable also had a large
number of missing values.
Sample
The analysis was conducted with 597 hospitals. The
original data set included 664 observations in at least
one of 14 files. After deleting four observations because
of duplicate identification numbers and six observations
that only appeared in one out of 14 files, there were 654
observations. Nineteen observations were excluded from
the analysis after examining inconsistencies across
variables. For example, eight hospitals had occupancy
rates greater than 200 percent. We did not have access to
the raw data, so in many cases it was not possible to
correct the data when an inconsistency was identified.
Thirty-eight observations were excluded, because data
were missing for one or more of the following variables:
cost (23 cases), admissions (16 cases), beds (12 cases),
inpatient days (10 cases) and region (two cases).
Recoding
Data were missing on outpatient visits for 78
hospitals, X-rays for 94 hospitals, operations for 88
hospitals, and lab tests for eight hospitals. These missing
values were recoded to zero based on two consistency
checks. First, the missing values were more likely to
occur at district hospitals where the services were less
likely to be available. Fifty-four out of 78 hospitals that
were missing data on outpatient visits were district
hospitals. Seventy out of 94 hospitals that were missing
data on X-rays were district hospitals. Second, the
frequency of a zero response for these variables was low,
suggesting that a zero was not entered as part of the skip
pattern of the data entry program. For example, there
were only four hospitals that reported zero outpatient
visits and three hospitals that reported zero X-ray
exams.
Estimation
The total variable cost Eq. (2) was estimated with
ordinary least squares using Stata, version 7. We
examined the estimates for multicollinearity and found
that it was not a problem.
The estimates were tested for heteroscedasticity using
the Cook and Weisberg test, also known as the Breusch–
Pagan test (Judge, Griffiths, Hill, & Lee, 1980), because
the residuals in cost function estimates may be
correlated with the independent variables. Heterosce-
dasticity does not affect the coefficient estimates, but can
cause incorrect inferences about their significance. All of
the tests showed heteroscedasticity. Consequently, ro-
bust standard errors were calculated using the Huber–
ARTICLE IN PRESSM. Weaver, A. Deolalikar / Social Science & Medicine 59 (2004) 199–208204
White method, and all inferences were based on these
standard errors.
The marginal costs in Eqs. (3) and (4) were calcu-
lated with the predicted value of cost. The predictions
were based on the smearing factor for a heterosce-
dastic normal distribution of the residuals (Manning,
1998). Statistics on kurtosis and skewness of the
residuals indicated that the normal distribution was
appropriate. Estimates with a logged dependent variable
require a special procedure for retransforming the
predicted value of the logged dependent variable back
into the original form; for example, transforming log
dollars into dollars. The predicted value must include a
smearing factor that transforms the geometric mean of
the logged dependent variable to the arithmetic mean of
the original variable. When the estimates are hetero-
scedastic, the smearing factor is different for each
observation and the predicted values of cost is calculated
with the mean of the predicted value rather than a
prediction based on sample means. The predicted value
of cost is
EðCÞ ¼ ea0þa1bedsef ðY ;X Þesmear=2; ð9Þ
where smear is the predicted value of the residuals-
squared when regressed on the independent variables in
the cost function.
Table 2
Variable cost function estimates with the coefficient for bed constrain
Variable Coefficient
Constant 19.76395
Central general 0.5584834
Central specialty 0.8276736
Provincial specialty 0.1850416
District �0.4532605
Other ministry �0.0034338
Beds (hundreds) 0.2161765
Admissions (thousands) 0.2022869
Admissions-squared �0.0074842
Admissions-cubed 0.0000814
Outpatients (thousands) 0.0022916
Admissions�outpatients �0.000095
Operations (thousands) 0.0151028
X-rays (thousands) 0.0006581
Lab tests (thousands) 0.0003646
Index of case complexity 0.0018437
Red river delta 0.2380451
North coast 0.3949181
Central coast 0.0422343
Central highland 0.1763819
South east 0.3149447
Mekong river delta 0.1001492
Sample size=597, adjusted R2=0.82, MSE=0.467, condition index=
Results
Descriptive statistics
The descriptive statistics in Table 1 show large
differences in the mean cost and number of beds across
categories. The mean cost ranged from $US 1.8 million
for central general hospitals to $US 0.5 million for
provincial general hospitals and $US 83,182 for district
hospitals. The mean number of beds ranged from 561 in
central general hospitals to 357 in provincial general
hospitals and 76 in district hospitals.
Despite the large size of the central general hospitals,
the majority of patients were treated at provincial
general and district hospitals. There were only 10 central
general and 18 central specialty hospitals compared to
76 provincial general, 74 provincial specialty and 385
district hospitals. Consequently, only 4 percent of
admissions were at central general and two percent at
central specialty hospitals, compared to 34 percent at
provincial general, 13 percent at provincial specialty and
43 percent at district hospitals.
The descriptive statistics also show clear differences in
the number of admissions and ALOS between specialty
and general hospitals. ALOS was 50 days at central
specialty and 36 days at provincial specialty hospitals
compared to 15 and 9 days at the central and provincial
general hospitals, respectively.
ed to be the same across categories of hospitals
Huber–White
standard error
T-statistic P-value
0.1165549 169.57 0.000
0.2465504 2.27 0.024
0.1474411 5.61 0.000
0.096631 1.91 0.056
0.0922304 �4.91 0.000
0.1264252 �0.03 0.978
0.0436749 4.95 0.000
0.0155873 12.98 0.000
0.0007868 �9.51 0.000
0.0000121 6.73 0.000
0.0004742 4.83 0.000
0.0000309 �3.07 0.002
0.0079839 1.89 0.059
0.0016786 0.39 0.659
0.004577 0.80 0.426
0.0018851 0.98 0.328
0.0732274 3.25 0.001
0.0636638 6.20 0.000
0.0629158 0.67 0.502
0.0744149 2.37 0.018
0.0882341 3.57 0.000
0.06135 1.63 0.103
38.84.
ARTICLE IN PRESSM. Weaver, A. Deolalikar / Social Science & Medicine 59 (2004) 199–208 205
Cost function estimates
Results for the first specification are reported in Table
2. The estimate has a single variable for beds and
dummy variables for each category of hospitals.
Provincial general hospitals were the omitted category
and serve as the reference group. As shown, the
coefficient for beds was large and significant indicating
that the hospitals were not at their long-run equilibrium.
The coefficients for the linear, squared and cubed terms
for admissions were of the expected sign and significant.
The coefficient for outpatient visits was positive and
significant. The coefficient for the interaction of admis-
sions and outpatient visits was negative and significant,
indicating of economies of scope. The coefficients for the
category of hospital showed significantly higher cost at
the central hospitals and significantly lower cost at the
district hospitals. There were significant differences
across regions, indicating higher cost in some of the
more populated regions like the Red River where Ha
Noi is located and the South East where Ho Chi Minh
City is located. Finally, the index of case complexity was
positive, but not significant in the estimates with robust
standard errors.
Marginal costs, SRVF, economies of scale and
economies of scope are reported in Table 3. As shown,
results for the total sample of hospitals obscure
important differences across categories of hospitals.
For the total sample the marginal cost per admission
was $US 34, but it ranged from $US 170 for central
general hospitals to $US 12 for district hospitals. For the
total sample the SRVF was 1.07 indicating roughly
constant returns, but the SRVF ranged from 0.87 for
central general hospitals to 1.44 for district hospitals and
1.73 for other ministry hospitals. For the total sample
there were modest economies of scope, but the measure
ranged from 0.77 for provincial general and 0.56 for
central general hospitals to almost zero for other
ministry hospitals.
The measure of economies of scale was problematic,
because it was negative for central general hospitals. The
problem could have been due to misspecification of the
Table 3
Marginal cost, SRVF and economies of scale and scope based on co
Total Central
general
Centr
specia
Marginal cost per admission
($US)
34.04 169.64 82.83
Marginal cost per outpatient visit
($US)
0.46 2.92 1.08
SRVF 1.07 0.87 1.31
Economies of scale 0.65 �0.30 0.60
Economies of scope 0.12 0.56 0.05
variable cost function or to the small number of central
general hospitals. A second specification was estimated
that added interaction terms for beds and the category
of hospital. A likelihood ratio test confirmed that these
unrestricted estimates were a better specification than
those with only a single variable for beds (w2 50.29,
p-value 0.000). A third estimate was performed with
additional interaction terms for admissions and the
category of hospitals to test whether or not the cost
function should be estimated in subsamples for each
category of hospital. A likelihood ratio test rejected the
specification with interaction terms for admissions and
the category of hospital (w2=1.53, p-value=0.91).
Results for the second specification are reported in
Table 4. As shown, the coefficient of the interaction term
for beds and the category of hospital was negative and
significant for central general hospitals. The sum of the
coefficients for beds and the interaction term for
beds� central general hospitals was �0.084, which
may indicate that these hospitals were operating at
close to a long-run equilibrium. The coefficient of the
interaction term was positive and significant for district
hospitals and other ministry hospitals. Other results
were similar to those reported in Table 2.
Marginal cost, SRVF, economies of scale, and
economies of scope for the second specification are
reported in Table 5. As shown, the measure of
economies of scale for central general hospitals was no
longer problematic. The measure was 1.09 for central
general and 1.05 for central specialty hospitals indicat-
ing roughly constant returns to scale. The measure was
well below one for both provincial general and specialty
hospitals indicating large diseconomies of scale. The
measure was 1.16 for district hospitals and 0.89 other
ministry hospitals indicating modest economies and
diseconomies of scale, respectively.
Overall, the pattern of results across categories of
hospitals was the same in both specifications. In both
Tables 3 and 5, the measure of SRVF was less than one
for central general hospitals, but greater than one and
relatively large for central specialty, provincial general,
district and other ministry hospitals. In both Tables 3
st function estimates with a single variable for beds
al
lty
Provincial
general
Provincial
specialty
District Other
ministry
15.74 59.91 11.59 17.81
0.39 0.69 0.15 0.23
1.51 0.89 1.44 1.73
0.21 0.48 1.18 1.38
0.77 0.19 0.05 0.01
ARTICLE IN PRESS
Table 4
Variable cost function estimates with different coefficients for beds and intercepts for each categories of hospital
Variable Coefficient Huber–White
standard error
T-statistic P-value
Constant 19.86426 0.1555143 127.73 0.000
Central general 2.080689 0.2701646 7.70 0.000
Central specialty 1.064031 0.2790802 3.81 0.000
Provincial specialty 0.1679008 0.1699075 0.99 0.323
District �0.5722532 0.1507081 �3.80 0.000
Other ministry �0.3635885 0.1958996 �1.86 0.064
Beds (hundreds) 0.2403785 0.0444017 5.41 0.000
Beds� central general �0.3245729 0.0469689 �6.91 0.000
Beds� central specialty �0.1046962 0.0966887 �1.08 0.270
Beds�provincial specialty �0.0050618 0.0500218 �0.10 0.919
Beds�district 0.214922 0.0887296 2.42 0.016
Beds�other ministry 0.4493576 0.1557589 2.88 0.004
Admissions (thousands) 0.1563481 0.0163716 9.55 0.000
Admissions-squared �0.0054898 0.0008111 �6.77 0.000
Admissions-cubed 0.0000556 0.0000119 4.68 0.000
Outpatients (thousands) 0.0020004 0.0004577 4.37 0.000
Admissions�outpatients �0.0000689 0.0000254 �2.71 0.007
Operations (thousands) 0.0115129 0.0084942 1.36 0.176
X-rays (thousands) 0.0020988 0.0014386 1.46 0.145
Lab tests (thousands) 0.0007165 0.0004229 1.69 0.091
Index of case complexity 0.0014695 0.0018985 0.77 0.439
Red river delta 0.1995971 0.717298 2.78 0.006
North coast 0.364869 0.0601444 6.07 0.000
Central coast 0.0207992 0.0596734 0.35 0.728
Central highland 0.1767505 0.0747761 2.36 0.018
South east 0.2657614 0.0920049 2.89 0.004
Mekong river delta 0.1117064 0.0605943 1.84 0.066
Sample size=597, adjusted R2=0.84, MSE=0.4498, condition index=53.94.
Table 5
Marginal cost, SRVF and economies of scale and scope based on cost function estimates with a coefficient for beds for each category of
hospitals
Central
general
Central
specialty
Provincial
general
Provincial
specialty
District Other
ministry
Marginal cost per admission ($US) 85.32 57.97 14.06 42.18 9.39 16.93
Marginal cost per outpatient visit
($US)
2.10 0.87 0.51 0.59 0.14 0.25
SRVF 0.74 1.52 1.40 1.06 1.77 2.12
Economies of scale 1.09 1.05 0.20 0.58 1.16 0.89
Economies of scope 0.48 0.05 0.61 0.18 0.05 0.01
M. Weaver, A. Deolalikar / Social Science & Medicine 59 (2004) 199–208206
and 5, there were large economies of scope for central
and provincial general hospitals compared to the small
economies of scope for central specialty, district and
other ministry hospitals. One notable difference between
the specifications was that the marginal cost per
admission was lower in Table 5 than in Table 3 for all
categories of hospitals and about half as much ($US 85
vs. $US 170) for central general hospitals.
Discussion
The large sample of public hospitals in Vietnam
allowed us to test for differences across categories of
hospitals in marginal costs, SRVF, economies of scale
and economies of scope. The results demonstrated that
inferences based on the total sample obscured funda-
mental differences across categories. Results differed
ARTICLE IN PRESSM. Weaver, A. Deolalikar / Social Science & Medicine 59 (2004) 199–208 207
across levels of hospitals and between general and
specialty hospitals. The pattern of results across
categories of hospitals persisted in two specifications of
the total variable cost function and with two different
measure of inpatient care.
Historically, public hospitals in Vietnam were allo-
cated primarily by administrative units rather than by a
market. In a system of public hospitals that follows an
administrative structure, returns to scale may depend on
the category of hospital in addition to the number of
beds and volume of output. Most previous research on
returns to scale has focused on the number of beds and
volume of output. For example, a recent review reported
consistent evidence of economies of scale for hospitals
with 100–200 beds, and diseconomies of scale for
hospitals with 300–600 beds (Sowden et al., 1996). In
contrast, in Vietnam the central general hospitals with a
mean of 561 beds exhibited constant returns to scale, as
did the central specialty hospitals with a mean of 226
beds. Provincial general hospitals with a mean of 357
beds and provincial specialty hospitals with a mean of
192 beds exhibited decreasing returns to scale. Among
the smaller hospitals, district hospitals with a mean of 76
beds exhibited increasing returns to scale, whereas other
ministry hospitals with a mean of 84 beds exhibited
decreasing returns to scale.
These results suggest that there were important
differences in managerial resources across categories of
hospitals. At the time of the hospital survey, there was
no justification for increasing the number of beds at
provincial hospitals with the managerial resources that
were available. It may be worthwhile to study hospital
management in different categories of hospitals to
identify problems at the provincial level and potential
solutions based on the experience of well-run hospitals.
It may also be worthwhile to explore ways to divide the
provincial hospitals into smaller units that would be
more appropriate for the managerial resources at that
level.
The results on SRVF also have policy implications.
SRVF were increasing for every category of hospital
with the exception of central general hospitals and
provincial specialty hospitals. These results imply that
there were gains to increasing admissions and outpatient
visits with the existing stock of beds. The increase in
admissions would require reducing the ALOS, because
the mean occupancy rates were high for all categories of
hospitals.
A rough comparison of marginal costs in Table 5 and
1995 hospital prices suggests that the price system
subsidized care at most categories of hospitals. Possible
exceptions were outpatient care at district hospitals and
inpatient care at specialty hospitals. The comparisons
were complicated by the fact that the price system was
by hospital type rather than hospital category (Govern-
ment of Vietnam, 1995). Central hospitals were gen-
erally in types one and two, district hospitals were
generally in types three and four, and provincial
hospitals were distributed throughout all four types.
The price of a general outpatient exam in types one and
two ranged from $US 0.14 to $US 0.27; the marginal
cost ranged from $US 2.10 for central general hospitals
to $US 0.51 for provincial general hospitals. The price in
types three and four ranged from $US 0.05 to $US 0.18;
the marginal cost of $US 0.14 for district hospitals fell
within that range.
For inpatient care, the rough comparison is based on
the marginal cost per admission in Table 5 divided by
the ALOS in Table 1. The price of an inpatient day at an
internal medicine department for types one and two
ranged from $US 0.54 to $US 0.91; the marginal
cost per admission/ALOS ranged from $US 5.86 for
central and $US 2.19 for provincial general hospitals to
$US 0.75 for central and $US 0.69 for provincial
specialty hospitals. The price for types three and four
ranged from $US 0.18 to $US 0.45, which was also less
than the marginal cost/ALOS of $US 1.03 for district
hospitals.
As mentioned in the Methods section, there was no
evidence that hospital reforms like bonuses for staff or
social insurance affected total variable cost. These data
were from an early stage of reforms in a health sector
that was undergoing rapid transformation. It may be
worthwhile to repeat the analysis with more recent data
by conducting a similar survey in this decade or
exploring data available through the Vietnamese health
information system.
As one of the first studies of hospitals in a developing
country with a large sample and a broad range of
hospital sizes, there were few studies in the literature
with which to compare our results. Barnum and Kutzin
(1993) was the only other study to calculate measures
for different levels of hospitals. They reported diseco-
nomies of scale and significant economies of scope for
their total sample of hospitals in China. When they
compared economies of scale and scope across levels
of hospitals, the diseconomies of scale were larger in
upper level hospitals than in middle and lower level
hospitals, and the economies of scope were smaller in
upper level hospitals than in middle and lower level
hospitals.
Despite the unprecedented quality of the Vietnamese
hospital data, the main limitation of this research was
that the estimates were performed with cross-section
rather than panel data. Carey (1997) and Arnould (2001)
estimated hospital cost functions with longitudinal data
from the United States and Wagstaff and Lopez (1996)
estimated them with longitudinal data from the Catalan
region of Spain. The longitudinal analysis adjusts for
unobserved individual effects of each hospital, such as
quality of services, unmeasured dimensions of case
complexity, and managerial ability. Carey (1997) com-
ARTICLE IN PRESSM. Weaver, A. Deolalikar / Social Science & Medicine 59 (2004) 199–208208
pared cross-section and longitudinal analyses and
rejected the null hypothesis of no correlation between
the coefficients estimated with cross-section data and
individual effects. The resulting bias in the cross-section
estimates affected the coefficients of the inpatient
variables and measures of economies of scale. Data on
the Vietnamese hospitals were available for 3 years, and
we will pursue longitudinal estimates of the hospital cost
function in future research.
Conclusion
In a system of public hospitals that followed an
administrative structure, the total variable cost function
differed significantly across categories of hospitals.
Economies of scale did not depend simply on the
number of beds and volume of output; large hospitals
in one category of hospital had constant returns to
scale, whereas smaller hospitals in another category had
large diseconomies of scale. Among the smaller hospi-
tals, district hospitals had modest economies of scale
and other ministry hospitals had modest diseconomies
of scale.
Acknowledgements
The authors would like to thank Drs. Vung, Vuu and
Trinh, and Lan Phuong Nguyen for providing informa-
tion about the health care system in Vietnam, Susan
Ettner for guidance on econometrics, and David
Rubanowice for work on the statistical analyses. They
are also grateful to two anonymous reviewers for
insightful comments. Partial funding for the analysis of
the Vietnamese hospital data was provided by the
Swedish International Development Agency (SIDA)
and the World Bank.
References
Aletras, V. H. (1999). A comparison of hospital scale effects in
short-run and long-run cost functions. Health Economics, 8,
521–530.
Anderson, D. L. (1980). A statistical cost function study of
public general hospitals in Kenya. Journal of Developing
Areas, 14(2), 223–235.
Arnould, R. (2001). The effect of hospital mergers and
acquisitions. An examination of cost and price outcomes.
Paper presented at the International Health Economics
Association meetings in York.
Barnum, H., & Kutzin, J. (1993). Public hospitals in developing
countries: Resource use, cost, and financing. Baltimore, MD:
Johns Hopkins University Press.
Bitran-Dicowsky, R., & Dunlop, D. W. (1993). The determi-
nants of hospital costs: An analysis of Ethiopia. In A. Mills,
& K. Lee (Eds.), Health economics research in developing
countries. New York: Oxford University Press.
Carey, K. (1997). A panel data design for estimation of hospital
cost functions. Review of Economics and Statistics, 79(3),
443–453.
Government of Vietnam, Ministry of Health/Ministry of
Finance/Ministry of Labor, Government Cost Committee.
(September 30, 1995). Interministrial circular No. 14 on
hospital user fees.
Grannemann, T. W., Brown, R. S., & Pauly, M. V. (1986).
Estimating hospital costs: A multiple output analysis.
Journal of Health Economics, 5, 107–127.
Jei Corporation. (2002). Daily Vietnamese Dong rate against
US Dollar, January 1, 1995 to August 2000. Available on
line at: http://www.jeico.co.kr/cnc57vtn.html. Accesssed on
July 5, 2002.
Judge, G. G., Griffiths, W. E., Hill, R. C., & Lee, T.-C. (1980).
The theory and practice of econometrics. New York: Wiley.
Le Duc Chinh. (undated). Hospital activities in Vietnam. Hanoi:
Ministry of Health of Vietnam.
Manning, W. G. (1998). The logged dependent variable,
heteroscedasticity, and the retransformation problem.
Journal of Health Economics, 17(3), 283–295.
Roemer, M. I., Moustafa, A. T., & Hopkins, C. E. (1968). A
proposed hospital quality index: Hospital death rates
adjusted for case severity. Health Services Research, 3,
96–118.
Sowden, A., Aletras, V., Place, M., Rice, N., Eastwood, A.,
Grilli, R., Ferguson, B., Posnett, J., Sheldon, T., & Sykes,
D. (1996). Hospital, volume and healthcare outcomes, costs
and patient access. Effective Health Care, 2, 1–16.
Vita, M. G. (1990). Exploring hospital production relationships
with flexible functional forms. Journal of Health Economics,
9(1), 1–20.
Wagstaff, A., & Lopez, G. (1996). Hospital costs in Catalonia:
A stochastic frontier analysis. Applied Economic Letters, 3,
471–474.
Weaver, M., & Rubanowice, D. (1999). Three studies of
hospital finance from the survey of hospitals in Vietnam.
Unpublished paper prepared for the World Bank’s review of
Vietnam’s health sector.
World Bank. (1993). World Development Report 1993: Investing
in health. New York: Oxford University Press for the World
Bank.
World Bank. (2001). Growing health: A review of Vietnam’s
health sector. Available online at http://www.worldbank.
org.vn/data pub/reports/bankk/rep25/index.htm. Accessed
on February 19, 2002.
Wouters, A. (1993). The cost and efficiency of public and
private health care facilities in Ogun State, Nigeria. Health
Economics, 2(1), 31–42.