report no. 45111-yf serbia baseline survey on cost and efﬁ

January 26, 2009

Document of the World Bank

Report N

o. 45111-YF

Serbia B

aseline Survey on Cost and Effi ciency in Prim

ary Health C

are Centers before Provider Paym

ent Reform

s

Report No. 45111-YF

SerbiaBaseline Survey on Cost and Effi ciency inPrimary Health Care Centers before ProviderPayment Reforms

Human Development Sector UnitEurope and Central Asia

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

Pub

lic D

iscl

osur

e A

utho

rized

SERBIA Baseline Survey PHC Centers

CONTENTS

Page

ACKN 0 W LED GEMENT S ............................................................................................... i

EXECUTIVE SUMMARY .............................................................................................. ii

Chapter 1 . INTRODUCTION ...................................................................................... 1 1.1

1.2

Background on Health Sector Reforms ................................................................. 1

Experience with Payment Reforms ....................................................................... 4

Chapter 2 . DATA AND METHODOLOGY .............................................................. 6 Data on Primary Health Care ................................................................................ 6

2.2 Outcome Measures ................................................................................................ 7 2.3 Analytical Methods ............................................................................................... 8

2.1

Chapter 3 . RESULTS PRIMARY HEALTH CARE ............................................... 10 3.1 Descriptive Analysis ............................................................................................ 10

General characteristics o f DZs ................................................................... 10 Expenditures in DZs ................................................................................. -12 Input use o f DZs ......................................................................................... 13 Outputs produced by DZs .......................................................................... 15 Revenue sources o f DZs ............................................................................ 17 Productivity o f DZs ................................................................................... -17

Summary o f Econometric Results ....................................................................... 18

Summary o f Baseline Performance Measures ..................................................... 18

3.1.1 3.1.2 3.1.3 3.1.4 3.1.5 3.1.6

3.2

3.3

Chapter 4 . CONCLUSIONS AND RECOMMENDATIONS ................................. 21

REFERENCES ................................................................................................................ 26

ANNEX ............................................................................................................................. 30 Annex 1 : Technical Annex: ........................................................................................... 3 0

2) Econometric Methods Used in this Analysis .................................................. 35 3) Results from Econometric Analysis ............................................................... 38 Results Production Function ..................................................................................... 3 9 Results Cost Function., .............................................................................................. 41 4) Model Building Process .................................................................................. 43

1) Review o f Studies on Production Functions and Cost Functions ................... 30

Annex Table 1 : List o f PHC Centers Surveyed. January . December 2007 ................. 6 1

Annex Table 2: Variables in the PHC Facility Survey. January - December 2007 ..... 64

Annex Table 3: Staffing in Primary Health Care Centers. January - December 2007 -65

Annex Table 4: Questionnaire for Primary Health Care Centers .................................. 66

Figures

Figure 1.1: Hospital beds per 100. 000 pop .......................................................................... 2

Figure 1.2: Hospital admits per 100 pop ............................................................................. 2

Figure 1.3 : Average length o f stay ...................................................................................... 2

Figure 1.4: THE in YO o f GDP ............................................................................................. 2

Figure A . 1 : Ranking o f DZs by Production Efficiency Score ........................................... 41

Figure A.2: Ranking o f DZs by Cost Efficiency Score ..................................................... 43

Figure A.3: Ranking o f DZs by Production Efficiency Score across Different Model

Specifications .................................................................................................................... -53

Figure A.4: Ranking o f DZs by Production Efficiency Score across Different Units o f

Analysis (Service Point) .................................................................................................... 54

Figure A.5: Ranking o f DZs by Cost Efficiency Score across Different Model

Specifications., ................................................................................................................... 60

Tables

Table 1.1 : Public facilities providing health services. in 2008 ............................................ 1

Table 2.1 : DZ Performance Measures Affected by Capitation Payment System ............... 8

Table 2.2: Descriptive variables and expected results o f production analysis .................... 9

Table 2.3: Descriptive variables and expected results in cost analysis ............................... 9

Table 3.1 : Description o f D Z Characteristics .................................................................... 11

Table 3.2: DZ Annual Expenditure on Inputs in 2007. in % o f total expenditures ........... 12

Table 3.3 : Distribution o f Staff in DZs. in percent of staff category ................................ 14

Table 3.4: Large equipment in working order. in % o f DZ category ................................ 14

Table 3.5: Utilization o f DZ Space, in percent o f total square meter ................................ 15

Table 3.6: Output o f DZs, number o f services .................................................................. 16

Table 3.7: DZ Referrals, number o f referrals and in percent of total visits ....................... 16

Table 3.8: DZ Sources o f Revenue, in percent o f total revenues ...................................... 17

Table 3.9: DZ Productivity, number o f visits in DZs ........................................................ 18

Table 3. 10. Baseline Values o f DZ Performance Measures ............................................. -19

Table A . 1 : Estimation o f the Stochastic Frontier Production Function ............................ 40

Table A.2: Determinants o f Inefficiency in DZ Production .............................................. 40

Table A.3: Estimation o f the Stochastic Frontier Cost Function ....................................... 42

Table A.4: Determinants o f DZ Cost Inefficiency ............................................................ 42

Table A.5 : Summary o f Alternative Models and Specifications Estimated ...................... 44

Table A.6. Likelihood Ratio Test Statistics for Alternative Specifications o f the

Traditional Production Function ........................................................................................ 46


Stochastic Production Frontier Model ............................................................................... 49

Table A.8. Estimates o f Inefficiency as a Share of Total Variance Across Alternative SF

Production Model Specifications ....................................................................................... 52


Traditional Cost Function .................................................................................................. 55

Table A. 10. Likelihood Ratio Test Statistics for Alternative Specifications o f the

Stochastic Frontier Cost Function ....... , ...... . , ...... . .... ,. . ..... .. .... , ..... , ...... .... ... . ...I..............I .... 57

ACKNOWLEDGEMENTS

This report was prepared by a team headed by Pia Schneider (Senior Economist, ECSHD), and comprising Cheryl Cashin (Consultant), Johannes Koettl (Junior Professional Officer, ECSHD), Ana Djordjevic (Consultant, World Bank Belgrade), and the research team working with CESID led by Predrag Djukic and Milos Mojsilovic. Cheryl Cashin was the main author, wrote the methodology, conducted and wrote the descriptive and econometric analysis and conclusions. The CESID researchers were responsible for data collection, data entry and management and organizing the final workshop. Johannes Koettl provided research assistance, and Ana Djordjevic and Hermina Vukovic logistical support in Belgrade. Sreypov Tep and Viktoria Lebedeva were responsible for the document processing. Pia Schneider developed the study design and finalized the report. Detailed comments and suggestions at various stages o f preparation were provided by the Minister o f Health o f Serbia H.E. Tomica Milosavljevid and the Deputy Minister o f Health Elizabet Pauvnovic; and by Abdo Yazbeck, Tamar Manuelyan Atinc, Armin Fidler, Julian Schweitzer and Mukesh Chawla (all World Bank). Thanks go to Adam Wagstaff, Lead Economist (DEC) and Jack Langenbrunner, Lead Economist (EASHD), who provided helpful comments on an earlier draft; and to Randy Ellis, Professor o f Economics, Boston University who was the peer reviewer o f this final report. This study was sponsored by the World Bank health sector strategy implementation program administered by Mukesh Chawla Sector Manager HDHNP.

The team gratefully acknowledges the collaboration provided by the Ministry o f Health, the Institute o f Public Health, and the Health Insurance Fund o f the Government o f Serbia. The team expresses i t s special gratitude to the staff working in all Primary Health Care facilities and hospitals who have participated in this survey and dedicated their time to fill in the questionnaires. Valuable feedback to the preliminary findings o f this report was provided by the participants o f a workshop in Belgrade on September 25,2008.

i

EXECUTIVE SUMMARY

1. The Serbian Ministry o f Health (MOH) and the Health Insurance Fund (HIF) are planning to change the provider payment from currently a line-item budget to (i) capitation payment in Primary Health Care (PHC) centers, and (ii) some form o f case-based payment such as diagnosis-related groups (DRGs) in hospitals. With this payment change the Government aims to set incentives to providers that will lead to a more efficient provision o f care and contribute to the sector’s financial sustainability.

2. The purpose o f this study i s to conduct a baseline survey on the cost and efficiency in Primary Health Care (PHC) Centers (Dom Zdravlja - DZ) in Serbia before the implementation o f the payment reforms. Data were collected in 147 DZs (see Annex Table 1). Results are used (i) to inform the payment reform and (ii) to establish a baseline on health sector performance including utilization, quality, cost and efficiency against which the impact o f the reforms can be assessed in a follow-up survey. Recommendations about the technical payment system design and capitation formula are beyond the scope o f this report and have been undertaken as a separate activity. This study was conducted with the support o f World Bank health sector strategy funds’.

3 . The current line-item budgets paid from the HIF to DZs and hospitals are based on the number o f staff who are allocated to health facilities according to their number o f beds. This creates an incentive to providers to use more staff and beds, but does not reward better productivity, quality o f care, or health outcomes. Under capitation the DZs will be paid, in advance, a pre-determined fixed rate to provide a defined set o f services for each individual enrolled with the DZ for a fixed period o f time. Capitation sets incentives to improve efficiency through reduced input use per patient, more output achieved with fewer inputs (e.g. more visits per physician), combining inputs more effectively (e.g. shifting some expenditures from staff and uti l i t ies to medicines and supplies), increasing preventive services, and providing fewer diagnostic services.

4. Under case-based payments such as DRGs, hospitals will be paid the average cost o f producing a “case” in an average hospital, which may be adjusted to account for regional economic conditions, and include indirect costs such as teaching and capital cost. A shift from line-item budgets to case-based payment in hospitals i s expected to lead to more inpatient admissions, shorter average length o f stay and higher patient turnover per bed, which may also increase hospital expenditures for the HIF.

5. During the past years, several donors have assisted the M O H and the HIF in provider payment and structural reforms. The World Bank, European Agency for Reconstruction (EAR), and the Red Cross in Kraljevo have proposed different

’ Healthy Development: The World Bank strategy for HNP results. April 24, 2007.

.. 11

capitation formulae for PHC. The MOH has not decided yet on the capitation formula and i s currently investing in the availability o f patient data, legal changes and institutional reforms, and the population registration with their preferred PHC providers. In 2007/8, the M O H with World Bank support has started preparatory work for DRG costing in six pi lot hospitals to estimate risk-distributions and costs. Future efforts will focus on information technologies, data management and analysis in hospitals and in the HIF, monitoring, evaluation and fine-tuning o f the case-mix and DRG rate. The MOH has conducted a human resources strategy ( M O H Serbia, 2005) and a health sector restructuring strategy (Sanigest, 2007). The human resource strategy guided the 9.5% reduction in the health workforce from 2004 until 2007. The M O H has closed 1,835 hospital beds since 2004; and based on the restructuring strategy plans to close additional hospital beds by 20 10. With the support o f the EAR and the World Bank, the MOH has strengthened management in health facilities to ensure that directors use their increased responsibility in decentralized health facilities to adjust their input factors such as staff and equipment*. The additional value added through this baseline survey i s to provide a comprehensive analysis o f the performance in PHC centers, against which any future reforms that affect provider behavior can be assessed.

6. The methodology used in the baseline survey includes descriptive analysis o f key performance measures in PHC centers, as well as an econometric analysis o f the current production and cost functions in PHC centers. The analysis aims to provide insight into the current level o f efficiency as wel l as the determinants o f the factors that influence efficiency. A technical annex contains the econometric literature review, methodology, analysis, results and describes the model building process.

7. The main finding from this baseline survey i s that DZs differ substantially in their efficiency. Although DZs are generally working with the same level o f staff, medical equipment and space, which are largely dictated by the system, they produce different levels o f output such as consultations etc. To some extent the level o f productivity in DZs may be affected by the age/gender structure o f the population, particularly by the number o f children in the DZ catchment area. There i s very little variation in the cost-efficiency o f DZs, because DZ expenditures are largely pre-determined as prices o f input factors (e.g. wages) are defined on a national level.

8. Additional findings show that expenditures in DZs are dominated by personnel costs (70% o f total cost). This i s at the expense o f medicines and supplies, which are also needed to improve the scope and quality o f DZ services. DZs are currently very constrained by their fixed costs and thus in their ability to improve cost efficiency, as their personnel costs are determined externally by the system. If personnel costs are excluded from capitation and the HIF continues to pay for

In addition, two recent Bank reports on decentralization and on the experience with provider payment reforms in Europe have informed the planned health sector reforms (World Bank, 2007; Schneider, 2007). 2

iii

staff based on a line-item budget, then only about 30 percent o f total DZ costs can be managed by DZs under capitation.

9. There are large areas o f unused space in DZs that are reducing DZ productivity. H a l f o f DZs have at least some large equipment but relatively few diagnostic tests appear to be performed. Thus, the productivity o f DZ staff, equipment and space can be improved by using equipment more often or reducing the number o f equipment, and reducing non-clinical space in DZs.

10. DZ mainly produce curative visits, and there appears to be an excessive number o f laboratory tests and injections; whereas relatively few preventive care services are provided. Referral rates for DZs are generally l o w which may be related to a similar l o w severity case-mix across DZs, but the current data do not allow determining whether patients are being referred more or less than necessary. DZs could become more productive by either increasing the number o f consultations and other outputs, or reducing the number o f staff without reducing the total number o f visits.

1 1. There i s currently unequal allocation o f public resources for primary care across DZs. This i s likely to be the result o f the way the line-item budget i s now defined based on the number o f staff and other factors. This inequality can be addressed by the capitation payment system if funding from the HIF, municipalities, and other government sources are pooled at a higher level and distributed to DZs through a unified capitated rate per enrolled individual.

12. Once capitation has been introduced, it may be expected that DZs will provide more preventive care visits to patients to reduce the need for more expensive diagnostic and curative visits; in addition, unnecessary laboratory tests and injections may also be reduced. Also, capitation may lead to higher referral rates to hospitals, as DZs have the incentive to reduce their cost, and hospitals paid by DRGs have an incentive to hospitalize more patients. Therefore, additional measures may be needed to make sure that capitation leads to prevent adverse effects on quality and access to care and on hospital spending.

13. The Government, in collaboration with health sector partners, has already started implementing several measures to prepare the sector for the planned provider payment reform, including:

(a) The Government i s currently reviewing proposals to adjust the capitation rate by coefficients for age and gender o f the enrolled population, and to include additional incentives for preventive services, such as a bonus payment to DZs who achieve an agreed level for childhood immunization or maternal care. Also, the Government i s considering geographic adjustment coefficients in the capitation payment to adjust for higher utility costs in DZs in mountainous areas.

iv

(b) The M O H has started a review o f clinical guidelines to ensure that these are compatible with the scope o f services that will be financed by capitation, and provide appropriate guidance to staff on laboratory tests, injections, other procedures, and referrals.

(c) The M O H and HIF in collaboration with donors have provided extensive management support to DZ managers such that they will be able to successfully respond to the incentives set by the new per capita payment system. This collaboration i s ongoing and includes management training and new accounting systems in al l DZs.

(d) The Government i s undertaking major investment in the data systems in DZs and HIF in collaboration with the World Bank loan and the EAR. The HIF has started to improve reporting and analysis o f key data related to DZ performance, including population size and demographic structure, services provided, and resource use.

14. Financial incentives set by the capitation payment system may not be enough to trigger a behavioral change among DZs that leads to more efficient care. Thus, other supporting policy changes and improved data and information systems may be required to realize the benefits o f the new payment system. Based on the conclusion from this baseline analysis, several additional steps could support the development o f capitation payment in DZs, prevent adverse effects in reaction to the financial incentives set by the payment, and improve the efficiency o f the sector. These measures could be implemented in a phased approach with Phase I focusing on the following seven steps.

Payment System Design

(a) Pool PHC funds from the HIF and other public sources, and pay al l DZs a unified capitation rate with appropriate adjustments for cost variations with the objective o f allocating primary care resource equally across DZ.

(b) Include salaries in the capitation amount and adjust related human resources policies and laws to give more flexibility to DZs to improve their cost efficiency.

(c) Specify in referral guidelines the scope o f services at the PHC level and appropriate referral pattern to prevent unnecessary referrals to higher-cost hospitals. In addition, the capitation system should include measures, such as open enrollment, a quality monitoring system, or outcome-based bonuses and penalties for unjustified referrals.

(d) While the Serbian health insurance law already sets the legal frame for copayments, the government may consider developing a revised cost-sharing policy as part o f overall provider payment reforms.

V

Management Issues

(e) Assess the regulations and constraints that may affect the ability o f DZs to manage their resources more efficiently. This will include regulations affecting the scope o f service at different levels o f care, public procurement laws, and the public labor law according to which public sector employees in DZs are s t i l l on the HIF payroll and planned centrally.

(0 Conduct an assessment o f essential medicines that are necessary for effective PHC, and consider the potential for limited financing o f essential medicines within the context o f capitation.

(g) Develop a cost-effective package o f medical equipment that should be available at the PHC level. Consider reducing the number o f equipments in DZs in the vicinity o f a hospital where patients could be referred. Examine whether more basic equipment i s available that may contribute more to DZ productivity (e.g. blood pressure cuffs and scales)

15. During Phase 11, additional attention could be given to the fol lowing three measures that arise from the conclusion o f this analysis:

Management Issues

(a) Consider reorganizing space in DZs, moving to smaller buildings, or redirecting excess space to other purposes. DZs could rent out non-clinical space to purposes such as private doctors or dentists, or to day-care centers for individuals who need supervision such as elderly or disabled individuals.

(b) Hospital payment reforms such as DRGs stimulate changes in hospital care such as shorter hospital stays that will be felt in other parts o f the health care system. As a result, DZs and community care as well as long-term care departments will have to be ready to provide a greater degree o f follow-up to patients who have been discharged from hospitals.

Data Systems

(c) Collect information on the disease profiles o f the populations served, outcomes, and overall access to essential medicines in the country. Develop quality and outcome measures that can be monitored by DZs themselves for internal management purposes, and at the system level by the HIF and MOH.

v i

CHAPTER 1. INTRODUCTION 1.1 BACKGROUND ON HEALTH SECTOR REFORMS

1. The Serbian Ministry o f Health (MOH) and the Health Insurance Fund (HIF) are planning to change the provider payment from currently a line-item budget to (i) capitation payment in Primary Health Care (PHC) centers, and (ii) some form o f case- based payment such as diagnosis-related groups (DRGs) in hospitals. The purpose o f this study i s to conduct a baseline survey on the cost and efficiency in Primary Health Care (PHC) Centers in Serbia before the implementation o f the payment reforms. Results can be used to inform the payment reform and to establish a baseline on health sector performance including utilization, quality, cost and efficiency against which the impact o f the reforms can be assessed in a follow-up survey. Recommendations about payment system design and capitation formula are beyond the scope o f this report and have been undertaken as a separate activity. This study was conducted with the support o f World Bank health sector strategy funds3.

2. Serbian PHC Centers (Dom Zdravlja - DZ) are organized either as a separate entity or as part o f the secondary care hospital - Zdravsteni centri. DZ provide basic primary services4. Patients in need for specialized primary care are referred to one o f the 19 specialized centers (Zavodi) or to one o f the 120 hospitals (Table 1.1). As part o f the decentralization strategy, al l DZ are becoming independent f rom hospitals.

Table 1.1: Public facilities providing health services, in 2008 Facility type Number of facilities Dom Zdravlja (Primary Health Care Center) 159 Zavodi (Specialized health center) 19 General Hospital (Opsta Bolnica) 37 Specialized Hospital (Specijalna Bolnica) 14 Single Specialty Clinic 23 Multi Specialty Institute 38 Clinical Hospital Center (Klinika Bolnica Centar) 5 Clinical Center (Klinika Centar) 3

Healthy Development: The World Bank strategy for HNP results. April 24, 2007. Including preventive health care, emergency care, general medicine, women and child health, visiting

nurse services, laboratory and other diagnostics. When other services are lacking in the area, the DZ also provides: dental care, occupational medicine, physical medicine, rehabilitation, ambulance transportation. When the catchment area for a DZ i s more than 20,000 people and it i s located more than 20 km from a General Hospital, the DZ also provides specialist services for: internal medicine, pneumo-physiology, ophthalmology, otolaryngology, psychiatry and cardiology.

3

1

3. In 2006, Serbia had relatively more hospital beds (Figure 1,1), lower inpatient admission rates (Figure 1.2) and longer average lengths o f hospital stays (ALOS) than neighboring countries (Figure 1.3). However, the MOH has taken several measures to reduce bed numbers and shorten ALOS in hospitals. Total health expenditure (THE) o f 8% o f GDP in 20055 was in l ine with neighboring countries (Figure 1.4).

Figure 1.1: Hospital beds per 100,000 pop Figure 1.2: Hospital admits per 100 pop

20 15 10

5 0

1 I ' Source WHO 2006. Source: WHO 2006. http://www.euro.who.int/hfadb http ://www.euro. who. intlhfadb

Figure 1.3: Average length of stay Figure 1.4: THE in YO of GDP 1

12 10

8 8 6 6 4 4 2 2 0 0

! :t

I I ' J

http://www.euro.who.int/hfadb Source: WHO 2006. Source: WHO 2005.

http ://w .euro, who .int/hfadb

4. The Serbian Government plans to change the provider payment system to set incentives to providers that will lead to a more efficient provision o f care. The current line-item budgets are based on the number o f staff who are allocated to health facil i t ies according to their number o f beds. This creates an incentive to providers to use more staff and beds, that define their budget, but does not reward improvement in productivity, quality o f care, or health outcomes.

5. Under capitation the payment to a DZ i s not linked to the inputs used or the volume o f services provided. Rather, the DZ will be paid, in advance, a pre-determined fixed

MOH Serbia: Development o f National Health Accounts in Serbia - Phase 111. August 3 1,2007 5

2

rate to provide a defined set o f services for each individual enrolled with the DZ for a fixed period o f time. Therefore, some financial risk i s shifted from the HIF to the DZ. If the DZ expenditures are greater than the capitation budget, it will be liable for this difference. If there are efficiency gains and costs are lower than the capitation budget, the DZ can retain and reinvest this surplus in the provision o f care which would be expected to lead to better health outcomes. Thus, capitation sets incentives to improve efficiency through reduced input use per patient, more output achieved with fewer inputs (e.g. more visits per physician), combining inputs more effectively (e.g. shifting some expenditures from staff and utilities to medicines and supplies), increasing preventive services, and providing fewer diagnostic services. This change in treatment behavior i s expected to lead to a more efficient input m i x (i.e. staff, drugs), fewer inputs and increased productivity, more efficient output mix, better quality o f care, and better health (Cashin et al. 2007).

6. Per capita payment may also create incentives for unintended consequences. For example, there may be an incentive for DZs with capitation payment to under-provide services and keep costs l o w or refer patients to specialists who are paid fee-for- service and hospitals. These negative outcomes may be prevented to some extent, with additional checks and balances in the system, such as open enrollment in DZs, a quality monitoring system, or outcome-based bonuses for better quality compliance, or even penalties if DZs skimp care on patients. Therefore, capitation rates are generally adjusted for age and gender o f the population registered with the PHC center, and geographic criteria. Some capitation formulas include quality-based components (e.g. cancer screening rates, immunization rates) to set incentives to providers to improve quality o f care and prevent adverse effects such as under- provision o f care.

7. Under case-based payments such as DRGs, hospitals are paid the average cost o f producing a “case” in an average hospital, which may be adjusted to account for regional economic conditions, and include indirect costs such as teaching and capital cost. A shift from line-item budgets to case-based payment in hospitals i s expected to lead to more inpatient admissions, shorter average length o f stay and higher patient turnover per bed, which may also lead to higher hospital expenditures for the HIF. However, as capitation i s expected to improve access to PHC services, it may be expected that hospitalization rates, particularly for conditions that can be prevented or managed at the DZ, will decline (Kozak et al, 2001).

8. During the past years, several donors have assisted the M O H and the HIF in provider payment and structural reforms. The World Bank, European Agency for Reconstruction (EAR), and the Red Cross in Kraljevo have proposed different capitation formulae for PHC. The M O H has not decided yet on the capitation formula and i s currently investing in the availability o f patient data, legal changes and institutional reforms, and the population registration with their preferred PHC providers. In 2007/8, the MOH with World Bank support has started preparatory

3

work for DRG costing in six pilot hospitals to estimate risk-distributions and costs. Future efforts wi l l focus on information technologies, data management and analysis in hospitals and in the HIF, monitoring, evaluation and fine-tuning o f the case-mix and DRG rate. The World Bank supported the M O H in conducting a human resources strategy (MOH Serbia, 2005) and a health sector restructuring strategy (Sanigest, 2007). The human resource strategy supported a 9.5% reduction in the health workforce from 2004 until 2007. The M O H has closed 1,835 hospital beds since 2004; and based on the restructuring strategy plans to close additional hospital beds by 2010. With the support o f the EAR and the World Bank, the M O H has strengthened management in health facilities to ensure that directors use their increased responsibility in decentralized health facilities to adjust their input factors such as staff and equipment. In addition, two recent Bank reports on decentralization and on the experience with provider payment reforms in Europe have informed the planned health sector reforms (World Bank, 2007; Schneider, 2007).

1.2 EXPERIENCE WITH PAYMENT REFORMS

9. The experience from other countries with introducing capitation payment in PHC and DRGs in hospitals (Schneider, 2007) i s being used in this analysis to describe how provider behavioral indicators presented in Table 2.1 may change in Serbia in response to the payment change. Corrective interventions by the M O H wi l l be proposed to prevent that any expected behavioral change by providers in response to the financial incentive set by the payment system, will have a negative effect on the provision and quality o f care.

10. Generally, provider payment reform tends to be undertaken in response to structural problems in the health sector that require a major reorientation o f overall financing and service delivery. For example, in 2002 New Zealand introduced PHC reforms, which included the formation o f new non-profit PHC entities funded through a per capita payment system, in order to address health disparities across socioeconomic groups that arose from the fee-for-service payment system (Hefford, 2005; M O H New Zealand 2001). In Costa Rica the PHC sector was reorganized into autonomous cooperatives paid by capitation in the early 1990s to address declining quality o f services, low morale among providers, and long waiting l i s ts for diagnostic and other services (Clark, 2002; Gauri et al, 2004). In the former Soviet Union, the historical neglect o f the PHC sector, over-specialized and fragmented care, an unsustainable hospital infrastructure, as well as l i t t le emphasis on prevention and health promotion brought about unprecedented declines in health status throughout the region early in the post-Soviet transition period (WHO, 200 1). Several post-Soviet countries embarked on comprehensive health financing and service delivery reforms, with the restructuring and strengthening o f PHC, supported by capitation payment (Borowitz et al, 1999; ZdravReform Program, 2000).

4

11. Provider payment reforms are rarely implemented in isolation, so it i s difficult or impossible to attribute changes in provider or system performance to the impact of the payment reform. Some studies do show that moving from a fee-for-service (FFS) payment system to capitation can have significant effects on costs and output. For example, when Ireland switched from FFS to a capitation system, physician visits declined by an estimated 20 percent. Moving from a line-item budget to capitation, as i s planned for Serbia, i s likely to have the greatest impact on efjciency through the flexibility that providers wi l l gain to change their mix of inputs. However, increasing provider flexibility wi l l require legal changes. Some evidence from former Soviet Central Asian republics shows that moving from l ine item budgets to capitation motivated providers to decrease expenditures from staff and utilities and invest in medicines and supplies, enabling PHC providers to increase the scope and quality o f their services (Cashin et al, 2007). The increased flexibility also allows PHC providers to change their output mix to emphasize preventive services, which may reduce costs for them by reducing follow-up visits. When a per capita payment system was combined with a quality monitoring system, PHC providers in the Karaganda region o f Kazakhstan shifted their services toward preventive services resulting in a 30 percent increase in prevention visits between 2001 and 2004. This shift was accompanied by a decline in the rate o f potentially avoidable hospital admissions for ulcers, asthma, and anemia (Cashin et al, 2007).

12. Case-based payments in hospitals such as DRGs generally result in more admissions and a shorter ALOS. During the past 20 years, the number o f hospitalizations increased markedly in countries with case-based payment, while it remained on a similar low level in Spain, Canada and the Netherlands, where physicians are paid a monthly salary independent o f their workload. Hospitals may also have an incentive to admit a patient who could be treated more efficiently in a D Z or day-surgery setting. In these cases, DRGs may conflict with overall expenditure controls by setting an incentive to increase the number o f hospitalized cases, resulting in higher hospital expenditures (Docteur et al. 2003). After the DRG payment was implemented in the U S Medicare system, ALOS in hospitals fel l by 15% in the f i rst three years; and f e l l as much as 24% for some diagnoses (Cashin, et al. 2005).

13. The rest o f this report i s organized i s follows. Chapter 2 presents the data and methodology used in this survey to evaluate the cost and efficiency performance in DZs. Results are presented and discussed in Chapter 3. Based on findings, Chapter 4 concludes and proposes several reform measures to support the effect o f the provider payment reform. The Annex contains additional information including a technical annex with an overview on the literature on cost and efficiency analysis, the econometric analysis, and the model building process; a l i s t o f health facilities included in the survey, summary results, and the questionnaire used to collect information.

5

CHAPTER 2. DATA AND METHODOLOGY 2.1 DATA ON PRIMARY HEALTH CARE

14. The M O H decided to survey al l DZs as this would provide a unique data base to establish a new inventory o f DZs following the recent organizational changes caused by the separation o f DZs from hospitals under decentralization. The data for this analysis come from a baseline survey o f revenues, expenditures, staffing patterns, and service provision in 147 o f 159 PHC centers in Serbia (see Annex Table 1). The survey does not include the 9 DZs in Kosovo6 and the 3 DZs who submitted incomplete questionnaires as they were undergoing organizational changes. A follow-up survey is expected to be conducted after the implementation o f the payment reforms in DZs.

15. Questionnaires collected existing routine data from DZs based o n monthly patient and facility registers. The questionnaires were developed by representatives from the MOH, CESID and the World Bank team and pilot-tested in February 2008 in three DZs7. The M O H sent questionnaires to all directors o f DZs in Serbia, and made reminder phone calls. Data collection took place over a six-week period in May-June 2008 and covers a time period o f twelve months from January - December 2007. Questionnaires were completed by staff in DZs in collaboration with field interviewers working for an independent local research firm CESID in Belgrade. Interviewers were trained, and visited all DZs at least once with follow-up visits to ensure data validity and completeness. CESID entered and cleaned data in Excel, which was transferred into STATA for analysis. Specific variables collected in the baseline survey are listed in Annex Table 2. The unit o f analysis i s the DZ.

16. Although the research team followed a consistent data collection approach, data quality may s t i l l be an issue in this study, and results should be interpreted with caution. The study relies on existing data, which are o f unknown quality, and there i s evidence o f some inconsistencies. For example, 70 percent o f DZs report more than 5 visits per year per person in the catchment population, and 25 percent o f DZs report more than 10 visits per person per year. This clearly raises concerns about inflated output data. In addition, a significant share o f DZs has, what appear to be inflated staff numbers, and for DZs that are s t i l l part o f a hospital complex, there was also a problem o f not being able to separate out expenditures and input use by DZs from those o f the parent health facility. These data quality issues raise concerns about the interpretation o f the results o f this study and also highlight the need for better data systems. However, results show that there i s sufficient variation in production efficiency levels and convergence on cost efficiency levels to conclude that poor data quality i s not entirely driving the results.

Under the UN Security Council Resolution 1244 (1999), Kosovo i s administered by United Nations Interim

BaEka Topola, Valjevo and VraEar. Administration Mission in Kosovo (UNMIK).

6

2.2 OUTCOME MEASURES

17. There are a variety o f methods for assessing the changes in provider performance in response to payment reforms. The experience from other countries shows that capitation payment i s expected to lead to a more efficient health sector. In the case o f health care providers, efficiency measurement assesses how wel l input factors such as labor, space, supplies, machines, and medicines are combined to produce different mixes o f services o f different levels o f quality. For example, if a provider increases quality o f care for the same or fewer resource inputs, it can be said that efficiency has improved. The ultimate output that health care providers can “produce” i s better health outcomes.

18. A set o f performance measures focused on efficiency are analyzed in this study, both through descriptive analysis and econometric analysis. Efficiency can either be measured from a productiodproductivity perspective, or from a cost perspective. We will focus on measuring efficiency, defined in multiple ways in an attempt to capture these dimensions. These measures and their expected change under the payment reform are summarized in Table 2.1, and will be examined in the follow-up study after the capitation payment system is introduced. In order to accurately compare efficiency, however, the severity o f the patient case-mix treated in DZs and quality o f output (e.g. health care services) should be accounted for, which i s not possible from the data available in this study.

19. In addition to the descriptive analysis, this study will include a technical annex that uses econometric methods to estimate efficiency in the production and cost-efficiency in PHC centers based on production and cost frontiers. The production frontier i s the maximum output possible with different combinations o f inputs, given the technology and other environmental constraints at the time. Technical efficiency i s achieved when maximum output i s produced from a given set o f inputs. Suppliers who produce levels o f output below the frontier are not technically efficient. The cost frontier i s the minimum cost that can be achieved to produce a given level o f output at given input prices. A provider i s considered cost efficient if output i s produced at the minimum achievable cost. Costs may exceed the cost frontier either because the output is below the potential maximum (technical inefficiency) or because the m i x o f inputs used is not optimal given current input prices, or a combination o f the two (Wagstaff and Barnum 1992). Reasons for technical inefficiency may include poor technology, organizational and managerial inefficiency, or problems with inputs. An organization can also improve efficiency by choosing the output m i x that maximizes profits or other desired outcomes (such as health outcomes), or allocative efficiency (Liu, 2000). Knowing whether health care providers are not achieving minimum costs because o f technical inefficiency or a sub-optimal input m i x may be important for policy purposes.

7

Table 2.1 : DZ Performance Measures Affected by Capitat ion Payment

# o f service points per km2

Total # o f visits per catchment population per month # o f preventive visits per catchment population.

Category Efficiency o f input use

outputs and reduce inputs per output. Increase or decrease, depending on the impact on total input costs. Providers have both incentive to reduce total outputs, and to increase preventive services. Increase as providers have incentive for prevention to avoid more expensive services.

Efficiency of output mix

% o f visits resulting in a referral to a specialist % o f visits resulting in a referral to a hospital 36 o f visits resulting in a referral to a specialist or hospital

Access

reducetotal output. May increase, as providers have incentive to reduce input costs. May increase, as providers have incentive to reduce input costs. May increase, as providers have incentive to reduce input costs

Potential unintended consequences

Average cost per visit Decrease as inputs become more productive, but may increase as there i s an incentive to

2.3 ANALYTICAL METHODS

20. This baseline study conducts both, descriptive analysis o f key DZ performance measures, as well as an econometric analysis o f the current production and cost

8

functions. The descriptive analysis o f D Z production wi l l analyze the following variables describing the level and type o f output by DZs, which can be expected to be affected by a change in the provider payment system (Table 2.2).

Variables

Total number of visits and/or services

Number o f visits per registered individual

Number o f visits per physician per month

Percent o f visits consultation, preventive, home visit

Expected result triggered by capitation payment

Total number of visits and services may decrease, since providers are not compensated for increasing output

The number of visits per enrolled individual may decrease, since providers are not compensated for increasing output.

Initially there i s an incentive for fewer overall visits, so the number o f visits per physician may decrease. But capitation also creates the incentive to reduce inputs or shift to lower cost inputs, including non- physician time. So, the number o f visits per physician per month may increase over the longer term.

The mix o f outputs may change, shifting to lower cost outputs. The share o f preventive visits may increase to reduce the need for more expensive diagnostic and curative visits.

21. The descriptive analysis o f costs in DZs wi l l analyze the following variables reflecting the level and mix o f inputs used by DZs. These variables may be proxy variables for efficient use o f inputs and can be expected to be affected by a change in the provider payment system (Table 2.3):

% o f revenue from patients

% o f expenditures on salaries, drugs, other

% o f staff physician, nurse, paramedical, admin

Table 2.3: Descriptive variables and expected results in cost analysis Variables I Expected result related after the introduction of capitation

If DZs improve efficiency o f input use the share o f payment from patients may decrease, depending on the payment definition

Greater flexibility in allocating expenditures across inputs may increase the share of DZ expenditures on drugs and supplies

Greater flexibility in allocating expenditures across inputs may increase the share of DZ staff expenditures on nurses and paramedical staff.

22. The econometric analysis in the Technical Annex provides insight into the current level o f D Z efficiency relative to the current capability o f the entire system, whether levels o f D Z efficiency vary, and the determinants o f variations in D Z efficiency.

9

CHAPTER 3. RESULTS PRIMARY HEALTH CARE 3.1 DESCRIPTIVE ANALYSIS

23. The descriptive analysis presents findings on the general characteristics o f DZs, input use, level o f output (services) produced, expenditure, revenue and DZs productivity. The analysis examines differences in inputs (e.g. staff), outputs (e.g. consultations) and productivity across a variety o f characteristics o f DZs including being rural or urban and stand-alone or being part o f a health center. It may be expected that results in a fol low up survey are different if DZs change their treatment behavior, as they respond to the new incentives created by the per capita payment system, or by the related organizational changes. Table 3.10 at the end o f this Chapter summarizes the main findings and can be used for comparison with a future survey.

3.1.1 General characteristics o f DZs

24. Table 3.1 summarizes the DZ characteristics. The DZs included in this study are evenly split between urban and rural (51 and 49 percent, respectively). The majority o f DZs are stand-alone clinics (70.8 percent), with the other 29.2 percent being part o f health centers. Stand-alone DZs are separate healthcare institutions and are not incorporated in health centers, which generally consist o f a hospital and one or more DZs. DZs typically have a central location with a number o f outposts. The average number o f outposts i s 1 1.3 per DZ. These service points are staffed by an average o f 8.5 physicians and 25.5 total medical personnel.

25. In terms o f ownership, 68.5 percent have been decentralized and are owned by the municipality, and 28.8 percent are owned by the M O H . At the time o f the baseline, no DZs are privately owned. Most DZs are rural stand-alone clinics owned by the municipality (65.8 percent).

26. The average distance to the nearest hospital i s 27.3 km, with one DZ 200 km from the nearest hospital. The DZs have on average 2.5 outpatient beds and 9.6 inpatient beds, but inpatient beds may have been counted that are actually part o f the DZ’s parent hospital. The average DZ catchment area i s 51 1.7 km2, ranging from 3 to 2,035 km2. The population density averages 381 people per km2, with a reported range o f 9 to1 8,806 inhabitants. Even taking outposts into consideration, some DZs cover a very large territory, which raises concerns about geographic access to PHC in Serbia, particularly in areas where there are no private GPs. For example, 62 DZs (42 percent) have a catchment area greater than 500 km2, and 13 DZs (9 percent) have catchment areas o f more than 1,000 km2. The average area per service point i s 84.8

10

km2 and exceeds 200 km2 for 11 DZs (7.5 percent), with one DZ reporting an area o f 1,500 km2 per service point. Furthermore, the number o f outposts i s not highly correlated with the size o f the catchment area or the catchment population. For example, one DZ with 22 outposts reports a catchment area o f only 7 km2, covering 55,543 people. Another DZ with 20 outposts has a catchment area o f 422 km2, covering 56,011 people.

27. The population served by DZs was reported both as the official catchment area, and as the number o f individuals registered with the DZ. The official catchment area i s more valid at this time, as the process o f registering individuals i s ongoing, and only a small percentage o f individuals had registered as o f the time o f the study. The catchment population ranges from 8,228 - 252,131 per DZ. The average population per service point i s 7,248, but the median i s 3,400 individuals and 80 percent o f DZs report a population per service point o f less than 6,000.

11

3.1.2 Expenditures in DZs

Percent o f total expenditure on

Personnel

Drugs

28.

29.

Mean per year (range) N=143

All DZs Rural Urban Stand-alone In Health Center

72.6 70.9 74.3 69.3 81.4 (0.27 - 100) (45.4 - 100) (0.27 - 91.5) (0.27 - 100) (59.0 - 91.5)

10.9 13.1 8.7 13.2 4.7

Total annual expenditure o f DZs in 2007 averaged 239 million Serbian Dinars (SD), or about U S $4.8 million. This represents a per capita expenditure o f 5,576 SD, or US$112. The per capita expenditure varies from 321 to 17,500 SD within DZs catchment area. The average per capita expenditure in the highest spending 25 percent o f DZs i s four times that o f the lowest spending 25 percent, indicating substantial inequality in the distribution o f primary care resources across DZs. Per capita expenditure i s not significantly related to whether the D Z i s urban or rural, or whether it i s stand-alone or part o f a health center. The average per visit expenditure i s 938 SD, or US$18.8. The average recurrent cost per visit i s 887 SD, or US $17.7, excluding expenditures on investment in equipment or infrastructure.

Supplies

DZs spend 72.6 percent o f total expenditures for personnel (Table 3.2). DZs that are part o f health centers spend a significantly higher share o f their total budget on personnel than stand-alone DZs, 8 1.4 percent vs. 69.3 percent, respectively.* This difference i s likely because in DZs that are part o f a health center, other inputs are provided or subsidized by the health center, and i t was not possible to attribute these costs to the DZ. Urban DZs devote a higher share o f their budget to staff than rural DZs, but this difference i s not statistically significant.

3.2 I 3.1 I 3.3 I 3.3 I 2.8

Utilities 3.5 I 3.4 I 3.6 I 3.3 I 4.0

Transport

Maintenance

Investment

Other

(0 - 11.5) (0-11.1) (0.21-11.5) (0-11.5) (0.89- 11.1) 2.1 2.4 1.9 2.2 2.0

(0 - 8.4) (0 - 6.3) (0.24 - 8.4) (0 - 8.4) (0.37 - 4.2) 3.3 3.1 3.5 3.4 3.1

(0 - 18.3) (0 - 8.8) (0 - 18.3) (0 - 18.3) (0 - 16.1) 4.9 4.6 5.1 5.6 2.9

4.4 4.1 4.6 5.3 1.9 (0 - 5 1 S ) (0 - 22.3) (0 - 5 1.5) (0 - 5 1 .5) (0 - 22.3)

(0 - 62.6) (0 - 38.0) (0 - 62.6) (0 - 62.6) (0 - 10.9)

* All comparisons o f means in the descriptive analysis are made using a simple two-tailed t-test.

12

30. DZs spend an average o f 10.9 percent o f their total budgets on drugs, and another 3.2 percent on supplies. Rural DZs allocate 16.1 percent o f their expenditures to medicines and supplies, while urban DZs allocate 12.1 percent. This difference i s statistically significant at the 10 percent level. Stand-alone DZs allocate 16.5 percent o f their expenditures to medicines and supplies, while DZs that are part o f a health center allocate only 7.6 percent. Again, this i s likely to reflect the difficulty in allocating health center expenditures to the DZs rather than an actual difference in patterns o f input use. However, if these DZs benefit from “free drugs” purchased by their parent hospital and they are about to be decentralized to independent DZs (World Bank, 2007), then they should budget for possibly future higher drug expenditures. About 9 percent o f D Z expenditures are allocated to other operating costs (utilities, transport, and maintenance), and 4.4 percent to other expenditures. Some DZs report a very high percentage o f expenditures as “other,” up to 62.6 percent, which indicates that they are not tracking their expenditures carefully.

3 1. DZs spent 10 million SD on investment per year, or about US$200,000, reflecting 4.9 percentage o f total D Z expenditure. Most o f the investment (8,275,206 SD) was for equipment, and 1,864,981 SD was for investment in infrastructure. There i s no difference in the level o f investment between rural and urban DZs, but stand-alone DZs allocate a significantly higher share o f total expenditures to investment (5.6 percent) than DZs that are part o f a health center (2.9 percent), suggesting that the latter may benefit from cross-subsidies.

3.1.3 Input use o f DZs

32. DZs have on average 265 total staff, with 61.8 physicians, 118 nurses, 3.5 paramedical staff, 13.7 administrative staff, and 36.0 technical staff. Per service point, the DZs have on average 35.8 total staff, including 8.5 physicians, 16.6 nurses, 0.43 paramedical staff, 2.7 administrative staff, and 6.6 technical staff. The ratio o f population to providers i s 195.5 per total staff, 782.4 per physician, and 389.7 per nurse. D Z staff i s typically paid based on the public wage regulations through the HIF, but DZs also hire an additional 7 percent o f staff from their own funds. Table 3.3 presents a summary o f D Z staffing patterns. A detailed description i s in Annex Table 3.

33. There appears to be some imbalance in the distribution o f staff across different DZ. On average, 25.4 percent o f DZ staff i s physicians, 50.7 percent are nurses, and 1, l percent i s paramedical staff (Table 3.3). The ratio o f nurses to physicians i s 2: 1. On average, more than 20 percent o f the staff i s non-medical, with 6.4 percent administrative and 16.5 percent technical. Rural DZs have a lower share o f physicians on staff, 24 percent vs. 26.8 percent, which significant. In addition, the ratio o f nurses to physicians i s higher in rural areas, 2.19 vs. 1.94 in urban areas, which significant. Rural DZs also have a significantly higher share o f administrative and technical staff. DZs that are part o f health centers have the same share o f

13

physicians, but a significantly higher share o f nurses than stand-alone DZs, and a significantly higher ratio o f nurses to physicians, with an average o f 2.20 vs. 2.02 in stand-alone DZs. DZs that are part o f a health center have a lower share o f administrative and technical staff, as health center staff may carry out many o f these functions.

staff Technical staff

(0 - 18.9) (0 - 18.9) (0 - 12.4) (2.6 - 18.9) (0 - 1 1.6) 16.5 17.3 15.6 16.8 15.7

(0-35.1) (0 - 26.2) (0 - 35.1) (5.5 - 26.2) (0-35.1)

34. Table 3.4 shows more than 44 percent o f DZs (64) report that they do not have an ultrasound or an X-ray machine in working order. Only one D Z reported having al l o f this equipment. Among those that do report having some equipment, the most common machine i s an ultrasound, with 52 percent o f DZs reporting at least one ultrasound machine, and 30 percent reporting two or more. Nearly half o f the DZs have an X-ray machine. Rural DZs have more equipment than urban DZs. Whereas 49.3 percent o f urban DZs report having no equipment in working order, only 35.7 percent o f rural DZs have no equipment that i s in working order. Results on equipment should be interpreted with caution as DZs who are s t i l l part o f a hospital complex may not have made a distinction between equipment that i s physically located in the hospital and in the DZ.

Equipment in working order

YO of DZ that have no ultrasound and no X-ray

Mean YO N=147

All DZs Rural Urban Stand-alone I n Health Center

43.5 37.5 49.3 42.3 46.5

At least 1

I machine

~ 52.4 I 58.3 I 46.7 I 52.9 1 51.2

14

At least 1 X-ray 46.3 I 48.6 I 44.0 I 45.2 I 48.8

35. Most o f DZ space i s used for non-clinical purposes or consultation rooms (Table 3.5). DZs report an average o f 5,797 square meters space, but space per service point averages 913 square meters. All but five DZs have a laboratory, and 66 percent o f DZs have a pharmacy. O n average, nearly ha l f o f the space in DZs i s not used for clinical purposes such as consultation room, laboratory, or pharmacy. Ha l f the DZs devote more than 50 percent o f space to non-clinical purposes. Rural DZs have a higher share o f space used for non-clinical purposes than urban DZs; and stand-alone DZs have a higher share than DZs in health centers.

Space Mean YO N=146

All DZs I Rural I Urban I Stand-alone 1 I n Health

Consultation Rooms Laboratory Pharmacy Other (non-clinical)

3.1.4 Outputs produced by DZs

~~~

43.1 42.8 43.3 41.9 45.8 3.8 3.4 4.1 3.3 4.8 3.5 4.2 2.9 4.3 1.7

46.8 49.2 44.5 47.9 44.2

36. The main output produced by a primary health care center i s consultations (Table 3.6). On average, DZs provide 409,194 consultations per year. About 9 percent o f consultations were preventive visits, and DZs report on average 4.4 percent o f total consultations as home visits. The shares o f preventive and home visits do not vary by DZ characteristics. Overall, emergency services make up only a small share o f services provided by DZs, less than 3 percent on average. Stand-alone DZs have a higher share o f emergency vis i ts (3.1 percent) than DZs that are part o f a health center (0.7 percent), but this difference i s not statistically significant.

37. DZs perform a large number o f laboratory tests, with an average o f 1.02 lab tests per consultation. Rural DZs perform fewer lab tests per consultation (0.89) than urban DZs (1.14), but this difference i s not statistically significant. DZs that are part o f a health center perform significantly more lab tests per consultation (1.45) than stand- alone DZs (0.84).

38. DZs also provide a large number o f injections, with more than hal f o f all consultations involving an injection. Rural DZs provide more injections per consultation than urban DZs, 0.78 vs. 0.34, which i s weakly significant. Stand-alone DZs have more injections per consultation than DZs that are part o f a health center, but this difference i s not statistically significant. However, there are very few diagnostic procedures per consultation, with an average o f 0.07, which does not differ very significantly by DZ characteristics (Table 3.6).

15

Table 3.6: OutDut o f DZs. number o f services

AllDZs I Rural

output

Urban I Stand-alone I I n Health

visits (% o’f total)

~ ~~~~~~~ ~~ ~

TOT #coriultations 1 409,194 1 293,946 1 500,029 I 390,594 I 407,502 Total # o f Dreventive I 36.545 I 21.793 I 43.887 I 34,341 I 29,979

(9.2%) I (9.0%) I (9.4%) I (9.5%) I (8.5%)

(% o f total) Total # o f emergency services (% o f total) Total # o f laboratory tests (per consultation) Total # o f diagnostic procedures (per consultation) Total # o f injections (per consultation)

5,524 1 8,699 I 22,077 I 16,063 I 14,22 1 (4.3 yo) (4.4%) (4 2%) (4.8%) (3.0%) 14,029 5,153 22,549 18,220 3,891

(2.4) (2.4) (2.4) (3.1) (0.7)

293,008 183,099 398,521 264,33 1 3 62,366 (1.02) (0.89) (1.14) (0.84) (1.45)

2 1,360 1 1,603 30,726 15,467 35,613 (0.07) (0.08) (0.06) (0.06) (0.09)

1 18,386 117,413 119,306 1 19,677 115,291 (0.56) (0.78) (0.34) (0.64) (0.38)

Referral

Total # o f referrals to specialists

(% o f total visits) Total # o f referrals

to a hospital (YO o f total visits)

Total # o f referrals (YO o f total visits)

39. Referral rates are relatively low among DZs but significantly higher in DZs that are part o f a health center (Table 3.7). Overall, 7.1 percent o f consultations result in a referral to a specialist, and 5.5 percent result in a referral to a hospital. The total mean referral rate i s 12.6 percent, which i s reasonable. Rural DZs have a higher rate o f referrals to hospitals (6.2 vs. 4.9) and total referrals (13.3 vs. 12.0), although these differences are not statistically significant. DZs that are part o f a health center have a significantly higher rate o f referrals to specialists than stand-alone DZs (8 .9 vs. 6.4), but there i s no significant difference in the rate o f referrals to hospitals or total referrals. Easy access to specialists in health centers may lead D Z providers that are s t i l l part o f a hospital complex, to more readily refer their patients.

Mean (per year) N=147

All DZs Rural Urban Stand-alone I n Health Center

19,795 14,664 25,066 17,924 24,3 18 (7.1) (7.1) (7.1) (6.4) (8.9)

(5.5) (6.2) (4.9) (5.4) (5.7)

37,245 24,181 49,786 34,148 44,735 (12.6) (13.3) (12.0) (11.8) (14.5)

17,450 9,5 18 25,066 16,224 20,4 18

16

3.1.5 Revenue sources of DZs

YO of DZ revenue from

40. The vast majority o f D Z revenues from the HIF, 84.2 percent o f total revenues on average, with very l i t t le variability across characteristics o f DZs (Table 3.8). The next most important source o f revenue for DZs i s from patients, with an average o f 7.6 percent. Only one DZ, however, reports a share o f revenue from patients greater than 22 percent. The percentage o f revenues from patients i s higher for DZs that are part o f a health center than for stand-alone DZs (9.1 vs. 7.1). Other sources o f revenue include municipalities, the MOH, Ministry o f Finance, and donors; though each contributes less than 3 percent to the total.

Mean per Year (range)

All DZs I Rural I Urban I Stand-alone I In Health

HIF

Patients

Municipality

Other government sources MOH

Donors

MOF

Center 85.4 84.2 86.5 84.3 88.4

7.6 7.5 7.7 7.1 9.1

2.1 2.6 1.6 2.7 0.53 '

(0 - 100) (0 - 100) (43.6 - 100) (9.9 - 100) (0 - 97.6)

(0 - 100) (0-100) (0-20.1) (0-21.4) (0 - 100)

(0 - 87.5) (0 - 87.5) (0 - 14.3) (0 - 87.5) (0 - 9.7) 0.95 1.2 0.75 1.3 0.01

(0 - 17.9) (0 - 17.9) (0 - 13.0) (0 - 17.9) (0 - 3.3) 0.88 1.5 0.30 1 .o 0.60

0.70 1.1 0.33 0.30 1.8

0.2 1 0.20 0.22 0.2 1 0.19

(0 - 44.3) (0 - 44.3) (0 - 3.0) (0 - 44.3) (0 - 16.4)

(0 -42.8) (0 - 42.8) (0 - 8.9) (0 - 8.9) (0 - 42.8)

(0 - 9.9) (0 - 9.9) (0 - 4.4) (0 - 9.9) (0 - 4.4)

3.1.6 Productivity o f DZs

41. The productivity o f DZs i s measured in terms o f the number o f visits per month per physician, per al l medical staff, and per individual in the catchment population per year (Table 3.9). Rural and stand-alone D Z report higher productivity. The average number o f visits per month per physician i s 561.2, and the average number o f visits per month per all medical staff i s 160.5. The average number o f visits per month per catchment population i s 0.70 (8 visits per person per year). This compares to 0.63 (7.6 visits per person per year) in the WHO Europe region. There are significant differences in productivity between rural and urban providers. Rural DZs report 636 visits per month per physician, whereas urban DZs report significantly less, only 488 visits per physicians. There are also more visits per catchment population in rural DZs, 0.81 vs. 0.60 in urban DZs. Stand-alone DZs report an average o f 529 visits per physician per month, and DZs that are part of a health center report 488 visits per physician per month. The number o f visits per catchment population in stand-alone DZs (0.77) i s significantly higher than in DZs that are part o f a health center (0.56).

17

DZ Characteristic

Mean (range)

# visits/physician/ # visitshedical # visitshegistered month staffhonth person/month

(N=122) (N=122) (N=125)

3.2 SUMMARY OF ECONOMETRIC RESULTS

All DZs

42. The main finding o f the econometric analysis presented in the Technical Annex i s that although DZs appear to be very constrained in their current ability to improve cost efficiency, there i s substantial variation in production efficiency. The efficiency scores o f more than half o f DZs (65%) are below 70 percent o f the frontier or the most efficient level. This shows that there i s clearly an opportunity to improve the production efficiency relative to the current capability o f the system, and to move al l DZs to a level o f higher efficiency. The productivity o f D Z health staff, equipment and space can be improved by using equipment more often or reducing the number o f equipment, and reducing non-clinical space in DZs. Less-efficient DZs would either have to increase the number o f consultations and other outputs, or reduce the number o f staff and medical equipment and rent out some o f the non-used space. There i s very l i t t le variation in the cost-efficiency o f DZs, because their expenditures are largely pre-determined. The quantity o f output does not give the full picture, however, and without being able to compare the quality of services or outcomes achieved, i t i s difficult to determine how much variation in efficiency i s actually taking place.

561.2 I 160.5 1 0.70

3.3 SUMMARY OF BASELINE PERFORMANCE MEASURES

Rural

Urban

43

(40.8 - 2377.4) (38.5 - 733.8) (0 - 4.32) 635.7 201.9 0.81

487.9 165.6 0.60 (40.8 - 2377.4) (1 1.9-733.8) (0 - 4.32)

Table 3.10 summarizes the above analysis to provide a baseline profile for the key efficiency performance measures o f D Z performance prior to the implementation o f per capita payment. Any follow-up analysis should compare results against these baseline values to identify changes in the behavior o f providers in response to the payment change.

Stand-alone

Part of a health center

18

(174.0 - 1274.9) (54.5 -493.0) (0 - 2.72) 582.8 194.6 0.77

488.0 155.4 0.56 (127.0 - 2377.4) (1 1.9-733.8) (0 - 4.32)

(40.8 - 155 1.7) (70.1-453.1) (0 - 2.78)

Category

Efficiency of input use

Efficiency o f output mix

’able 3.10: Baseline Values 01 Performance Measure

% of total expenditure on staff

% of total expenditure on medicine and supplies

% of total expenditure on utilities

% of total expenditure not categorized (“other”)

# visits per physiciadmonth

# visits per medical stafumonth

Populatiodphysician

Populatiodnurse

Marginal productivity of inputs (% change in output per 1 % change in units of input): Physicians

Nurses

Paramedical staff

Space

Equipment

Marginal cost of outputs (change in total expenditure per 1 unit change in output):

Consultations

Laboratory tests

Other diagnostic tests

% of total visits that are preventive

# of lab tests per consultation

# of diagnostic procedures per consultation

# of injections per consultation

12 Performance Measures Average Baseline Value

(Standard Deviation) 72.6% (14.8) 10.9% (12.5)

3.5% (1.8) 4.4% (6.9) 561

(375) 161

(123)

390 (1 04)

0.820 (0.26)

0.264 (0.290)

(but not significantly differentfrom 0 )

0.001 (0.028)

(but not significantly diyerentfrom 0)

-0.182 (0,124)

(but not significantly differentfrom 0)

0.016 (0.049)

(buf nof significantly differentfrom 0 )

0.003 SD (0.0007)

0.00 SD (0.008)

0.00 SD (0.003)

9.2% (8.8)

1.02 (0.99) 0.07

(0.14)

0.56 (1.39)

19

Category Performance Measure

Access km’ per service point

Total # of visits per catchment population per month # of preventive visits per catchment

average cost per visit

Potential unintended population per year consequences

YO o f visits resulting in a referral to a specialist

% of visits resulting in a referral to a hospital

referral to a specialist or hospital) Total referral rate (YO o f visits resulting in a

20

Average Baseline Value (Standard Deviation)

84.8 (163.8)

0.70 (0.60) 0.61

(0.54)

887 SD (1 128)

7.1% (0.61)

(0.90)

(1.1)

5 .5%

12.6%

CHAPTER 4. CONCLUSIONS AND RECOMMENDATIONS

44. The purpose o f this analysis was two-fold. Firstly, results provide a baseline on the PHC performance before the introduction o f capitation payment against which a follow up survey can be compared. Secondly, findings highlight opportunities to improve the efficiency o f DZs and support the introduction o f capitation payment. The main finding i s that although DZs are generally working with the same level o f staff, medical equipment and space, which are largely dictated by the system, they vary in the level o f output (e.g. consultations) they are able to achieve. That i s some DZs are substantially more efficient than others. These variations in output cannot be explained by whether the DZs are rural or urban, stand-alone or part o f a health center, or by the size o f the catchment area, the population density, or the distance to the nearest hospital.

45. The three main conclusions o f the baseline analysis can be summarized as follows:

(a) There i s room for DZs to manage their resources including staff, space and equipment, better, which may be stimulated by the financial incentives set through a per capita payment system. be enough to trigger a behavioral change that leads to more efficient care. Thus, other supporting policy changes and improved data and information systems may be required to realize the benefits o f the new payment system.

However, financial incentives may not

(b) There i s substantial inequality in per capita expenditures across DZs based on the current line-item budget. The new per capita payment system must contribute to redistributing resources more fairly and based on the number o f individuals enrolled, in order to achieve the goals o f improving access to services.

(c) A per capita payment system may give DZs more flexibility, but only if salaries are included under capitation and human resources policies are eased, which may require legal changes. DZs are currently very constrained in their ability to improve cost efficiency, as 70% o f their costs are for staffing and determined externally by the system, whereas a relatively small portion i s spent on medicines and supplies in DZs.

46. DZs can become more efficient by reducing the number o f staff and space, without having to reduce the total number o f visits. The finding that there are too many

21

47.

physicians i s supported by a very l o w population-to-physician ratio o f only 782 which i s much lower than in the WHO’S Europe region with 3,500 people per primary care p h y ~ i c i a n . ~ There are large areas o f unused space in DZs and DZ managers may not necessarily be aware about the cost o f space, as they do not have to pay rent o f space. DZs with more equipment are not showing a higher productivity.

DZ mainly produce curative visits, and there appears to be a high number o f laboratory tests and injections. Under capitation it may be expected that the share of preventive visits will increase and unnecessary laboratory tests and injections may be reduced. Stil l , there may be a need to revisit clinical guidelines to specify more clearly when laboratory tests and injections are necessary. Referral rates for DZs are generally low, though they may increase once capitation has been introduced; however, based on the current data it i s not possible to determine whether patients are being referred more or less than necessary.

48. Some DZs report a very high percentage o f expenditures as “other,” which indicates that they are not tracking their expenditures carefully, and highlighting the need for better accounting and resource management.

49. The Government has already started implementing several reform measures in collaboration with donors to prepare the health sector for the planned provider payment reforms. These measures include:

(a) The Government i s currently reviewing proposals to adjust the capitation rate by coefficients for age and gender o f the enrolled population, and to include additional incentives for preventive services, such as a bonus payment to DZs who achieve an agreed level for childhood immunization or maternal care. Also, the Government i s considering geographic adjustment in the capitation payment to adjust for higher utility costs in DZs in mountainous areas.

(b) The MOH has started a review o f clinical guidelines to ensure that they are compatible with the scope o f services that will be financed by capitation, and provide appropriate guidance on laboratory tests, injections, other procedures, and referrals.

(c) The M O H and HIF in collaboration with donors have provided extensive management support to DZ managers such that they will be able to successfully respond to the new capitation payment system. This collaboration i s ongoing and includes management training and investment in new accounting systems in al l DZs.

(d) The World Bank loan and the EAR are assisting the Government in major investment in the data systems in DZs and HIF. The HIF has started to

WHO European Health For All Database httr,://data.euro.who.int/hfadb/ accessed July 2008. 9

22

improve reporting and analysis o f key data related to DZ performance, including population size and demographic structure, services provided, and resource use.

50. Financial incentives set by the capitation payment system may not be enough to trigger a behavioral change among DZs that leads to more efficient care. Thus, other supporting policy changes and improved data and information systems may be required to realize the benefits o f the new payment system. Based on the conclusion from this baseline analysis, several additional steps could support the development o f capitation payment in DZs, prevent adverse effects in reaction to the financial incentives set by the payment, and improve the efficiency o f the sector. These measures could be implemented in a phased approach with Phase I focusing on the following administrative measures to complement the payment reforms.

Payment System Design

(a) Improve the equality o f primary care resource allocation by pooling PHC funds from public sources and paying all DZs a unified per capita rate with appropriate adjustments for cost variations. There i s currently unequal allocation o f public resources for primary care across DZs. This i s likely to be the result o f the way the line-item budget i s now defined based on the number o f staff and other factors. This inequality can be addressed by the per capita payment system if funding from the HIF, municipalities, and other government sources are pooled at a higher level and distributed to DZs through a unified capitated rate.

(b) Include salaries in the capitation amount and adjust related human resources policies and laws to give more flexibility to DZs to improve their cost efficiency.

(c) Explore measures to reduce the unintended consequences o f the per capita payment system. Capitation may create an incentive for DZs to under-provide services or refer patients to specialists and hospitals. To prevent unnecessary referrals to higher-cost providers, referral guidelines should specify the scope o f services at the PHC level and to appropriate referral patterns. These outcomes should be monitored and the payment system should include measures to counteract these incentives, such as open enrollment, a quality monitoring system, or outcome-based bonuses and penalties for unjustified referrals.

(d) Develop an appropriate cost-sharing policy. This study shows that DZ revenues from patients are currently 8 percent o f total DZ revenue. While the Serbian health insurance law already sets the legal frame for co-payments, the government may consider developing a cost-sharing policy as part o f overall provider payment reform that facilitates access to primary health care, i s

23

linked to essential drug reimbursement, and encourages appropriate use o f services at the different care levels.

Management Issues

5 1. DZs are currently very constrained by their fixed costs. If the HIF continues to pay for staff based on a line-item budget, then more than 70 percent o f the DZ budgets will continue to be fixed for salaries, and only about 30 percent o f total costs can be managed by DZs. In order to be effective at improving cost efficiency, DZs must have greater flexibility to shift expenditures f rom salaries and uti l i t ies toward more drugs and supplies; and from higher cost staff (physicians) to lower cost staff (nurses and paramedical staff).

(e) Assess the regulations and constraints that may affect the ability o f DZs to manage their resources more efficiently, including regulations affecting the scope o f service o f different levels o f medical personal, public procurement laws, and the public labor law according to which public sector employees in DZs are s t i l l on the HIF payroll and planned centrally.

(f) Conduct an assessment o f access to essential medicines that are necessary for effective PHC , and consider the potential for limited financing o f essential medicines within the context o f capitation. Collect information on the disease profiles o f the populations served, outcomes, and overall access to essential medicines in the country.

(g) Develop a cost-effective package o f medical equipment that should be available at the PHC level. Consider reducing the number o f equipments in DZs in the vicinity o f a hospital to where patients could be referred. Conduct a functional assessment o f the utilization o f equipment, including whether more basic equipment i s available that may contribute more to DZ productivity (e.g. blood pressure cuffs and scales)

52. Phase 2 could give additional attention to the fol lowing three administrative measures to complement the payment reform that also arise f rom the conclusion o f this analysis :

Management Issues

(a) Consider reorganizing space in DZs, moving to smaller buildings, or redirecting excess space to other purposes. DZs could rent out non-clinical space to purposes such as private doctors or dentists, or to day-care centers for individuals who need supervision such as elderly or disabled individuals.

24

(b) Hospital payment reforms such as DRGs stimulate changes in hospital care such as shorter hospital stays that will be fe l t in other parts o f the health care system. As a result, DZs and community care as well as long-term care departments will have to be ready to provide a greater degree o f follow-up to patients who have been discharged from hospitals.

Data Systems

53. There are some limitations o f the current analysis, which provide some guidance for where data improvements may be necessary to accurately monitor the effects o f capitation payment.

(c) Collect information on the disease profiles o f the populations served, outcomes, and overall access to essential medicines in the country. Develop quality and outcome measures that can be monitored by DZs themselves for internal management purposes, and at the system level by the HIF and M O H .

54. To evaluate the impact o f capitation over time a longitudinal analysis will be completed to assess after the payment reforms whether the reforms lead to an increase in productivity and a reduction in cost inefficiency. The study design will depend on the approach to implementing the per capita payment system and will, to the extent possible, exploit the phased-in implementation o f the new payment system. Better and more differentiated data could produce more reliable estimates o f marginal products and costs, which would allow additional analysis, such as whether the optimal m i x o f DZ inputs i s being used.

55. The follow-up survey will have to isolate the payment effect from the effect o f other variables that also may have changed after the payment change was implemented. For example, in the case o f measuring efficiency gains from implementing capitation payment, i t i s possible that there are also changes in such contextual factors as the availability o f medicines or changes in the supply o f physicians in DZs as a result o f the Government’s rightsizing plan, which would also affect efficiency. In addition, some DZs may be selected for participating in the new payment system earlier in the implementation process which may be related to factors that are also related to efficiency, such as their reputation for historically superior performance, or already having better data collection.

25

REFERENCES Ashenfelter, 0. and Card, D. (1 984). Using the longitudinal structure o f earnings to estimate the effect o f training programs. NBER Working Paper Series. Working Paper No. 1489.

Battese, G. and Coelli, T. (1995). A model for technical inefficiency effects in a stochastic frontier production function for panel data. Empirical Economics 20: 325- 332.

Borowitz M., O’Dougherty S., Cashin, C., Hafner, G., Samidjiyski, J., VanDevelde, C., and McEuen, M. (1 999). The Kazakhstan Country Program. Abt Associates Inc. USAID-Funded ZdravReform Program.

Cashin, O’Dougherty, et al. (2007). The design and implementation o f per capita payment systems in the context o f primary health care-centered health systems development in low- and middle-income countries. Abt Associates Inc. ZdravPlus Program. Almaty, Kazakhstan.

Clark, M. 2002. Health sector reform in Costa Rica: reinforcing a public system. Prepared for the Woodrow Wilson Center Workshops on the Politics o f Education and

DeFelice L and Bradford W D (1 997). Relative inefficiencies in production between solo and group practice physicians Health Economics (6)5: 455-465.

ECHSD Concept Note: Baseline study to evaluate the impact o f provider payment reforms on performance in health facilities in Serbia. December 18, 2007.

Fleetcroft, R. 2006. The relationship between prescribing expenditure and quality in primary care: an observational study. British Journal of General Practice 56(529): 61 3- 619.

Gauri, V., Cercone, J., and Briceno, R (2004). Separating financing from provision: evidence from 10 years o f partnership with health cooperatives in Costa Rica. Health Policy and Planning 19(5): 292-301.

26

Gaynor, M. and Pauly, M. (1 990). Compensation and productive efficiency in partnerships: evidence from medical group practice. The Journal of Political Economy 98: 544-573.

Gerdtham, U.-G., Lothgren, M., Tambour, M., and Rehnberg, C. 1999. Internal markets and health care efficiency: a multiple-output stochastic frontier analysis. Health Economics 8: 15 1 - 164.

Giannakas, K. et al. 2003. On the choice o f functional form in stochastic frontier modeling. Empirical Economics 28: 75-1 00.

Grannemann, T., Brown, R., and Pauly, M. 1986. Estimating hospital costs. Journal of Health Economic 5: 107-127.

Greb, S., Delnoij, D., and Groenewegen, P. (2006). Managing primary care behavior through payment systems and financial incentives. In Saltman, R., Rico, A,, and Boerma, W. Primary Care in the Driver’s Seat? Organizational reform in European primary care. Berkshire, England: Open University Press. pp. 184-200.

Gruber, J. (1 994). The incidence o f mandated maternity benefits. American Economic Review 84(3): 622-64 1,

Health Action International/WHO. 2008. Measuring medicine prices, availability, affordability and price components. Second edition http://www. haiweb.org/medicineprices/.

Hefford, M. (2005). Reducing health disparities through primary care reform: the New Zealand experiment. Health Policy 72: 9-23

Hollingsworth, B (2003). Non-parametric and parametric applications measuring efficiency in health care. Health Care Management Science 6(4) 203-2 18.

Hollingsworth, B. 2008. The measurement o f efficiency and productivity o f health care delivery. Health Economics 1 7: 1 1 07- 1 128

Jacobs, R., Smith, P., and Street, A. ( 2006). Measuring Efficiency in Health Care.: Analytic Techniques and Health Policy. New York: Cambridge University Press.

Johnson, S. and Lahiri, K. 1992. A panel data analysis o f productive efficiency in freestanding health clinics. Empirical Economics 17: 141 -1 5 1.

Kozak, L., Hall, M., and Owings, M. (2001). Trends in avoidable hospitalizations, 1980- 1998. Health Affairs 20(2): 225-232.

27

Kumar, R. 1999. Research Methodology: A Step-By-Step Guide for Beginners. London: SAGE Publications.

Liu, X. (2000). Improving the allocative efficiency o f health services. Global Program for Evidence for Health Policy. Geneva: World Health Organization.

M O H Serbia and David Cochrane: Serbia Human Resource Strategy for the Health Sector. 2005.

Ministry o f Health o f New Zealand. (2001). The Primary Health Care Strategy. Wellington, New Zealand.

Ministry o f Health o f the Kyrgyz Republic. 2008. Mid-Term Review Report: Manas Taalimi Health Sector Strategy. Bishkek, Kyrgyzstan.

Nordyke, RJ (2002). Determinants o f PHC productivity and resource utilization: a comparison o f public and private physicians in Macedonia. Health Policy 60( 1): 67-96.

Puig-Junoy J and Ortun V. 2004. Cost efficiency in primary care contracting: a stochastic frontier cost function approach. Health Economics (1 3): 1 149-1 165

Rosenman R and Friesner D. 2004. Scope and scale inefficiencies in physician practices Health Economics (13) 11: 1091-1 116.

Rosenman R, Siddharthan K, and Ahern M. 1997. Output Efficiency o f Health Maintenance Organizations in Florida Health Economics (6)3 : 295-302.

Ruggiero, J. 1999. Efficiency estimation and error te rm decomposition in the stochastic frontier model. European Journal of Operational Research 1 15(3): 555-563

Sanigest. Health Sector Restructuring Strategy: Network Plan. Serbia. Republic of Serbia. May 2007.

Suraratdech, C. and Okunade, A. 2006. Measuring operational efficiency in a health care system: a case study from Thailand. Health Policy 77: 2-23.

Schneider, P. (2007). Provider payment reforms: lessons from Europe and America for South Eastern Europe. Health, Nutrition and Population Discussion Paper. The World Bank: Washington, D.C.

Street A. 2003. How much confidence should we place in efficiency estimates? Health Economics (1 2) 1 1 : 895-907.

28

Wagstaff, A. and Lopez, G. 1996. Hospital costs in Catalonia: a stochastic frontier analysis. Applied Economic Letters 3: 471 -474.

Wagstaff, A. 1989. Estimating efficiency in the hospital sector: a comparison o f three statistical cost frontier models. Applied Economics 2 1 : 659-672.

Wagstaff, A. and Barnum, H. 1992. Hospital cost functions for developing countries. Population, Health and Nutrition Working Paper 1044. Washington, D.C.: The World Bank.

Weaver, M. and Deolalikar, A. 2004. Economies o f scale and scope in Vietnamese hospitals. Social Science and Medicine 59: 199-208.

Weissman, J. S., Gatsonis, C., and Epstein, A. M. 1992. Rates o f Avoidable Hospitalization by Insurance Status in Massachusetts and Maryland. JAMA. 268( 17): 23 8 8-23 94.

World Bank (2007). Serbia Decentralization and Local Service Delivery. Chapter 4. PREM. Europe and Central Asia.

World Health Organization (200 1). Basic Health Indicators Database. World Health Organization web site: http://www.who.int.

Zavras, A,, Tsakos, G., Economou, G., and Kyriopoulos, J. (2002). Using DEA to evaluate efficiency and formulate policy within a Greek national primary health care network. Journal of Medical Systems 26(4): 285-292.

ZdravReform Program (2000). Health reform initiatives in Central Asia: ZdravReform Program final report. ZdravReform Program: Almaty, Kazakhsta

29

ANNEX

ANNEX 1: TECHNICAL ANNEX":

1) Review of Studies on Production Functions and Cost Functions

56.

57.

58.

The concept o f efficiency measurement i s concerned with assessing the competence with which inputs are converted into outputs (Jacobs et al, 2006). The literature shows a m i x o f the production function and cost function approaches to analyzing health provider efficiency, and different units o f analysis (individual physician, provider organization, and purchaser). This review provides a summary o f approaches to empirical specification and estimation o f health provider production and cost functions.

Different measures o f output produced by health care providers are found in the literature. Most studies focusing on ambulatory care use, some variation o f services (usually visits) per clinic or physician per unit o f time. One study defined output as health, measured as the number o f enrollees in a primary care practice (Rosenman et al, 1997), and one study defined output as the size and composition o f the covered population, number o f visits, and quality o f care as measured by the percentage o f the population for which specific targets were achieved, including vaccination targets and coverage by a healthy chi ld program, and performance o f physicians on a simple knowledge test (Puig-Junoy, 2004).

Inputs are measured differently in cost and production functions. In cost function analysis, the use o f inputs i s summarized in a single measure o f the total cost o f the provider, or the total amount spent o n al l inputs combined. The literature varies substantially in how health care inputs are defined in production functions. One study specifies the number o f visits per physician per week (output) as a function o f the number o f hours per week worked by the physician, the number o f nursing hours per physician, and the number o f administrative hours per physician (DeFelice, 1997). Another study specifies inputs as physician time per visit, indicator variables for the availability o f basic medical equipment, and the number o f nurses per physician (Nordyke 2002).

Definition and Studies on Production Functions

59. Health provider production i s typically analyzed (modeled) as a joint ly determined process o f workload (output) and input utilization (cost) (Nordyke, 2002). The production function describes the relationship between the total output possible and

Cheryl Cashin PhD, wrote this technical annex I O

30

different levels and combinations o f inputs, such as labor and capital, as well as external factors that may affect total output (control variables).

60. Adding control variables to the production function recognizes that the production o f health care also embodies behavioral responses by providers, rather than simply a technical relationship between inputs and outputs (Gaynor and Pauly, 1990). These control variables may include physician characteristics, patient case-mix, and market characteristics (e.g., ownership, competition, prof i t status) and often reflect the policy variables o f interest, such as the provider payment method. Adding these variables allows defining a framework for estimating the effect o f individual providers or clinic characteristics on efficiency, as well as the change o f a policy variable (e.g. new per capita payment method) on provider input and/or output choices (Johnson and Lahiri, 1992).

The production function may be described as: Y = Y(L, K, 0, X), where:

Y L K 0 X

61

62

= level o f output (e.g. number o f patients) = amount o f labor input used = amount o f capital input used = amount o f other inputs used = other variables that can be expected to affect total output

The functional form o f the production function describes how inputs are expected to relate to output. For example, a Cobb-Douglas functional form i s used when it i s expected that output will change in direct proportion to changes in inputs. The translog production model i s flexible and often used to study how different inputs are substituted for each other in the production o f health care services. In one study o f the health care system o f Thailand, the authors used a translog production model to examine the production efficiency o f the health care system at the macro level (Suraratdech, 2006). The authors examined the productivity o f and substitutions between physicians, pharmacists, nurse, and beds. They found that although nurses and beds make a positive contribution to output (live births per thousand populations), a higher number o f physicians and pharmacists i s negatively associated with improved outcomes.

One study in Macedonia estimated a physician-level production function as three joint ly determined choices: output as patient visits per physician per week, inputs as time spent/patients and medical equipment/patient (Nordyke, 2002). The control variables included an indicator for whether the provider was public or private to test the impact o f facility ownership on efficiency. Results showed that private physicians had higher output (productivity) and chose a different input m i x (more eauiDment relative to labor) than Dublic sector Dhvsicians.

1 1 I ,

31

Definition and Studies on Cost Functions

63. The cost function describes the relationship between the total input use (cost) and the different levels o f output produced.

The cost function may be described as: C = C(Y, X), where:

C =the total cost Y = level o f output X = other variables that can be expected to affect total costs

64. Most studies estimate traditional cost functions that capture the multi-product nature o f health services production. Grannemann, Brown and Pauly (1 986) provided the basis for much o f the subsequent work on estimating cost functions for health care providers. The authors specified a cost function for hospitals as a function o f multiple outputs (inpatient days by type, discharges by type, outpatient visits, emergency department visits), input prices, and other factors that may be expected to shift the cost function (e.g. the sources o f revenue o f the hospital, the age structure o f the physician staff, form o f ownership o f the hospital, teaching affiliation, and geographic region). The objective o f their study i s to estimate the marginal cost o f production o f different outputs, recognizing that the level o f some inputs cannot change easily in the short run, but other inputs will vary with the level o f output. The marginal cost o f each type o f output i s expressed as a fraction o f total cost, varying with the level o f output. The authors found that hospital stays in pediatrics, surgery, obstetrics and gynecology, and other specialties are somewhat more costly than cases in general and internal medicine, even controlling for length o f stay. The authors also found that the source o f payment for the case, Medicaid and Medicare, significantly affected cost, with Medicaid inpatient days more costly than Medicare days.

65. In a study o f hospital costs in Vietnam, the authors use a variation o f Grannemann’s multiple-product cost function. A cost function i s specified in which hospital variable costs are a function o f input prices, the number o f hospital beds as a proxy for hospital capital stock, and the level o f output o f multiple products, including inpatient days and outpatient visits (Weaver, 2004). Interaction terms between the different products were included to estimate economies o f scope, or efficiencies that may be achieved by producing multiple products. The authors found large variations in the marginal cost o f hospital admissions across categories o f hospitals, from US$12 per admission in district hospitals, to US$170 per admission in central general hospitals, and modest economies o f scope.

66. Wagstaff and Barnum (1992) conducted a survey o f techniques available for analyzing hospital costs to answer specific pol icy questions, including (i) are hospitals over-capitalized? (ii) are hospitals inefficient? (iii) should hospitals specialize or provide a broad range o f services? and (iv) are there too many hospitals? The authors concluded that although the application o f current methods to hospital cost analysis in developing countries did not lead to clear guidance for

32

answering the policy questions, there were some methodological conclusions. First, the estimation o f short-run cost functions i s preferred to long-run cost functions, because hospitals may not be employing their long-run equilibrium quantities o f capital. Second, because o f the difficulty o f separating variable from fixed costs, it may be preferable to estimate short-run total cost equations. Third, since hospitals may be technically and allocatively inefficient, frontier models should be used. Fourth, to avoid too many restrictive assumptions about cost structures, flexible functional forms should be specified for cost functions. Finally, because economies o f scale are a long-run phenomenon, the estimation o f short-run cost functions may not fully reflect economies o f scale without calculation o f the optimal capital stock.

Definition and Studies on Stochastic Frontier Analysis

67. A widely accepted method for estimating health care production and cost functions i s stochastic frontier analysis (Puig-Junoy, 2004; Rosenmann, 2004; Street, 2003; Gerdtham, 1999; DeFelice, 1997; Battese 1995; Gaynor, 1990). This method estimates production and cost functions using ordinary least squares (OLS) regression methods. The least squares error term i s decomposed into two components: the random, “white noise” component, and the component attributable to inefficiency. The maximum production or minimum cost “frontier” i s estimated from the observation with the minimum error term, and the inefficiency o f the other observations i s estimated as the deviation from this minimum, excluding the

The SF method generates an overall efficiency value and random noise error ranking o f individual units based on how they perform relative to this value (Puig- Junoy and Ortun, 2004).

11,12,13.

68. Several studies use the stochastic frontier production approach to analyze provider response to reimbursement and compensation incentives. In one study using the stochastic production frontier approach to analyze provider production, the authors show that output-based reimbursement improved efficiency o f health care providers in Swedish counties (Gerdtham et al, 1999). In another study, the production frontier approach combined with a behavioral production function showed that productivity-based compensation affects the quantity produced by individual physicians but not technical efficiency for U.S. medical groups, because the increased output i s produced by also increasing inputs (Gaynor, 1990).

69. Two approaches to the stochastic frontier cost (SFC) model are found in the literature on health services costs. In the first approach, a cost function i s specified as a function o f single or multiple outputs, input prices, with other variables included that may shift the cost function. Average inefficiency i s estimated from the decomposed error term (Wagstaff, 1996; DeFelice, 1997; Street, 2003). The

Battese, G. and Coelli, T. (1995). A model for technical inefficiency effects in a stochastic frontier

Wagstaff, A. and Lopez, G. 1996. Hospital costs in Catalonia: A stochastic frontier analysis. Applied production function for panel data. Empirical Economics 20: 325-332

Economic Letters 3: 471-474. l3 DeFelice, L. and Bradford, W. 1997. Relative inefficiencies in production between solo and group practice physicians Health Economics (6)5: 455-465.

12

33

other approach i s to specify the cost function only as a function o f output and input prices, with a separate function estimating the inefficiency te rm that includes the cost shifters as explanatory variables for inefficiency, the parameters o f which are estimated simultaneously with the cost function (Battese, 1995; Puig-Junoy, 2004).

70. In a study o f hospital costs in Catalonia, the authors apply the SFC approach as an extension o f Granneman’s cost function model (Wagstaff, 1996). A cost function i s specified with a flexible functional form, with hospital costs a function o f ambulatory visits, emergency cases, and inpatient discharges. Variables are included that may be expected to shift the cost function, including teaching status, a proxy for case-mix complexity, and several variables reflecting the technology o f the hospital. The authors estimate the marginal cost o f producing each o f the three types o f hospital output, which as expected i s lowest for ambulatory visits and highest for inpatient cases. The authors also find complementarity between ambulatory visits and emergency cases, which i s reflected in modest economies o f scope. The inefficiency estimate from the model shows that the hospitals were on average operating 58 percent above the minimum cost frontier. One study used the SFC approach to estimate the average inefficiency o f hospitals in the U.K. Department o f Health (Street, 2003). The author found that the average mean level o f efficiency o f 90 percent estimated by the SFC model would have been underestimated by a traditional cost function, which estimated the mean level o f efficiency to be 69 percent.

7 1 . A SFC model also was used to estimate differences in efficiency between solo and group practice PHC physicians in the US (DeFelice 1997). The authors incorporated variables that affect physician effort, including physician experience and non-practice income. The results suggest that whether PHC providers are organized in solo or group practice does not have a significant effect on efficiency.

72. A different study in Catalonia used a SFC approach to measure the efficiency o f contracting out primary health care (Puig-Junoy, 2004). The authors found that there was no efficiency gain from contracting out PHC services instead o f direct public service provision. The cost function showed that the purchaser’s average contract with contracted-out PHC providers was 19.3% above the minimum cost frontier for a given level o f output, while the average contract with publicly managed PHC providers was only 7.6% above the minimum cost frontier.

73. The stochastic frontier approach i s widely used in efficiency analysis, but i t has several limitations. First, the approach requires strong assumptions about the distribution o f the error terms in order to disentangle inefficiency from statistical noise (Wagstaff 1989). The inefficiency error term i s often modeled as a half- normal distribution, which i s done in the present study, but this i s somewhat arbitrary. The SF approach has also been criticized for relying on the proper decomposition o f the error term into inefficiency and noise, which may be

34

74.

problematic (Ruggiero, 1 999),14 and being very sensitive to the specification o f the functional form (Giannakas et al., 2003; Hollingsworth, 2008).’57’6 A further limitation o f the stochastic frontier model i s that i t can only estimate the degree to which each observation deviates from the best production that i s observed, which i tse l f may also embody inefficiency (DeFelice and Bradford, 1997).

Another approach to efficiency analysis i s data envelope analysis (DEA) (Rosenman, et al, 1997). Data envelop analysis i s a linear programming method that derives the efficiency frontier strictly from the data, without the economic theory foundation or statistical approach o f stochastic frontier analysis. The observations with the highest ratios o f output to input are considered efficient, and the efficiency frontier i s constructed by joining these observations in input-output space (Jacobs et al, 2006). The DEA approach has been used extensively to analyze health care provider efficiency in a wide variety o f settings (Hollingsworth, 2003; Zavras et al, 2002). I t may be concluded from the literature that stochastic frontier approach i s preferred, however, because it i s a parametric method and therefore accommodates both statistical error as well as error attributable to inefficiency, which DEA does not. That is, the best practice frontier i s not strictly defined by the “highest” outputlinput pairs in the data, as observations are permitted to be somewhat above or below best practices due to measurement error and other random noise (DeFelice, 1997). In addition, SFC analysis can incorporate factors related to behavior (not just input-output mix), such as incentives, which allow an analysis o f provider response to policy changes.

2) Econometric Methods Used in this Analysis”

75. The econometric analysis presented in this study will use a SF model to estimate both the production and cost functions. The SF models will estimate the deviation o f each DZ from the best performance (maximum output or minimum cost) observed in the sample. This deviation i s considered to be caused by both inefficiency and random factors. A one-period production and cost functions will be estimated using the year 2007 as the baseline year. Although output variables are reported by month, and a time-series analysis could be possible, the input variables are reported for the entire year o f 2007, so they do not vary in the data.

Ruggiero, J. 1999. Efficiency estimation and error term decomposition in the stochastic frontier model. European Journal of Operational Research 1 15(3): 555-563. l5 Giannakas, K. et al. 2003. On the choice o f functional form in stochastic frontier modeling. Empirical Economics 28: 75-100.

Hollingsworth, B. 2008. The measurement of efficiency and productivity of health care delivery. Health Economics 17: 1107-1 128.

The complexity and power o f the econometric analysis i s limited by the content and quality o f the data. For example, analyzing both the production function and the cost function can have the added advantage o f shedding light on the sources of inefficiency (Wagstaff and Barnum 1996). If input prices are known, it i s possible to measure the excess costs to a provider caused by combining inputs in the wrong proportion using the marginal product estimates and input prices. Because the data do not allow disaggregation of expenditures, and therefore prices, across important input categories (e.g. physicians vs. nurses), this level o f analysis i s not possible.

14

16

17

35

76. The econometric analysis o f the production function for DZs will be completed in two stages. First, a stochastic production frontier function i s estimated to provide estimates o f the technical efficiency in the use o f the inputs by DZs in producing the main output o f consultations." Variables reflecting the proportion o f visits for children, adult women, and adults over 60 are added to control for the case-mix o f types o f visits to DZs. The first stage production function i s specified as a Cobb- Douglas production function as follow^:'^

InY, =p, lnL,, + p2 lnL,, + p3 lnL,, + p, In Space, +p, InEquipment, + PJnPropchild, + p,lnPropfem, + p,lnPropsenior[ +(E, + u t ) (1)

where,

lnY, 1nLp 1nLNi 1nLPAi InSpace, InEquipment, InPropchild, InPropfem, InPropsenior,

= = = = = = = = =

The natural log o f the total number o f visits per year in DZ iZo The natural log o f the total number physician full-time equivalents in DZ i The natural log o f the total number nurse full-time equivalents in DZ i The natural log o f the total number paramedical full-time equivalents in DZ i The natural log o f the total space in square meters in DZ i The natural log o f the total ndinber o f machines in DZ i The natural log o f the proportion o f visits by children 0-1 5 years in DZ i The natural log o f the proportion o f visits by females 16-59 years in DZ i The natural log o f the proportion o f visits by adults 60 and above in DZ i

El = independent, identically distributed random error for DZ i u, = non-negative unobservable random variable associated with technical

inefficiency for DZi that i s assumed to fo l low a half-normal distribution

The pj Cj=1-5) coefficients generated by the least squares estimation o f equation (1) will provide an estimate o f the productivity o f each o f the inputs including labor, equipment, space etc. The estimated value o f Vi provides information on the level o f production inefficiency o f provider i. The level o f inefficiency may be calculated as the ratio o f frontier maximum output to the observed output for provider i."

77. Second, an equation will be specified to determine the factors that explain differences in inefficiency across DZs. 22323 A variable will be included for the percentage o f revenue the provider receives through capitation, although this will be

'' Production functions were estimated to include two other outputs, laboratory tests and diagnostic tests, but no input variables were found to be statistically significant determinants o f these outputs.

A translog specification o f the production function was tested, because i t provides a general flexible functional form. A likelihood ratio test failed to reject the hypothesis that the second order input arameters are equal to zero, so the Cobb-Douglas function was used as the final specification.

19

A Cobb-Douglas specification o f the production function wi l l be used, so the empirical estimate wil l be based on the natural logarithm o f the output (visits) and inputs (full-time equivalents for physicians, nurses, and paramedical staff). 2 ' Puig-Junoy and Ortun 2004.

Battese and Coelli, 1995. Puig-Junoy and Ortim 2004.

22

23

36

zero for all providers in the baseline analysis, and reach a higher value in the follow-up survey.

u, = 4 capitation, + $2Private, + $3 Standalone, + $4 Rural, + $5 Radius, + $6Density, + $7Distance, + $.$taffRed+m, (2)

where,

Capitation, Private, Standalone, Rural, Radius, Density, Distance, StaffReduct,,

Percentage of the revenue of DZ i paid by capitation An indicator variable (011) for whether DZ i i s private An indicator variable (0/1) for whether DZ i part of a hospital/health center An indicator variable (0/1) for whether DZ i i s located in a rural area The radius of the catchment area o f DZ i The population density o f the catchment area o f DZ i The distance from DZ i to the nearest hospital % o f total staff of DZ i reduced between 2007 and period t due to the Ministry human resources policy a random variable

The $1 coefficients generated by the least squares estimation o f equation (2) will provide an estimate o f the impact o f each o f the external factors on DZs' production inefficiency. The production function i s estimated using thefrontier function in STATA 9.0.

78. The econometric analysis o f DZ costs will use a SF cost function analysis. The total expenditure o f the DZ i s specified as a function o f the level o f output across different service categories. Typically input prices are also included when estimating the cost function, but i t can be assumed that all DZs face the same input prices. There may be some variation by urbadrural areas, which will be captured by including a variable for rural DZ. Again, variables reflecting the proportion o f v is i ts for children, adult women, and adults over 60 are added to control for the case mix o f types o f visits to DZs. The empirical SF cost function i s specified as follows:24

where, lnC, InY,, 1nYLt lnYDi In Propchi Id, InPropfem, InPropsenior, " 1

natural log o f the total annual expenditures of DZ i natural log of the annual total number o f consultations in DZ i natural log of the annual total number o f laboratory tests in DZ i natural log o f the annual total number o f diagnostic services in DZ i The natural log of the proportion of visits by children 0-1 5 years in DZ i The natural log of the proportion o f visits by females 16-59 years in DZ i The natural log o f the proportion o f visits by adults 60 and above in DZ i independent, identically distributed random error for provider i

24 Again, a more flexible functional form was used initially following that o f Wagstaff and Lopez 1996, but the parameters on second and third order interaction terms between inputs were not found to differ significantly from zero, so the simpler specification was used in the final presentation o f results.

37

non-negative unobservable random variable associated with cost inefficiency for provider i that i s assumed to follow a half-normal distribution

- -

The a, Cj=1-3) coefficients generated by the least squares estimation o f equation (3) wil l provide an estimate o f the average marginal cost o f producing an additional unit o f each different type o f output. The estimated value o f u, provides information on the level o f cost inefficiency o f DZ i.

79. A separate function will be specified to model the effects o f independent variables on inefficiency. This step o f the analysis will allow the effect o f the Ministry’s human resources pol icy on input use to be separated from the affect o f the new per capita payment system.

v, = 0, Capitation, + @,Private, + 6, Standalone, + 0, Rural, +e3 Radius, + B,Density, + B,Distance, + @,StaffRed,, + P, (4)

Where al l variables are specified as in equation (2) above. The ’? coefficients generated by the least squares estimation o f equation (4) wil l provide an estimate o f the impact o f each o f the external factors on the cost inefficiency o f DZs. The cost function i s estimated using thefrontier function in STATA 9.0 with the cost option.

80. The production and cost functions are used to produce estimates o f marginal productivity o f inputs, or the contribution to output o f employing an additional unit o f the input; and the marginal cost o f output, or the contribution to total variable costs o f producing an additional unit o f the output.

81. The analysis o f equations (1) through (4) i s completed using STATA statistical software. The fol lowing specific analysis i s completed. In the follow-up study, the impact o f capitation payment on efficiency can be estimated from this basic model.

- - -

A comparison o f the productivity o f different inputs Analysis o f the contribution to total costs o f different outputs Estimate o f the average level o f production and cost inefficiency o f DZs, and ranking o f DZs by efficiency Analysis o f the impact o f external market factors, including the Ministry’s restructuring policy, on efficiency

-

3) Results from Econometric Analysis

82. The main objectives o f the regression analysis presented in this Section are to: (1) identify the marginal productivity o f key inputs, and the marginal cost o f production for the main outputs o f the PHC centers (DZs); and (2) measure and provide a

38

baseline for assessing changes in provider performance, in particular efficiency, after the implementation o f per capita payment. In order to meet both o f these objectives, a stochastic frontier (SF) model was chosen for this analysis over the estimation o f traditional production and cost functions, which do not provide the opportunity to compare the efficiency o f different units. Results on the model building process are presented in Section 4.

Results Production Function

83.

84.

85.

The DZ production function i s estimated to assess the baseline productivity o f DZ inputs, and the current level o f inefficiency relative to what i s achievable in the current system. The estimated coefficients presented in Table A.l reflect the productivity o f different inputs in producing DZ consultation^^^. Only the number o f physicians i s significantly associated with increased output. The related coefficient can be interpreted as a one percent increase in the number o f physicians in DZs leads to a 0.82 percent increase in the number o f visits, which i s significant at the one-percent level (p<O.Ol). Nurses and paramedical staff do not have statistically significant relationship to the level o f output o f DZs. The space in square meters o f DZs has a negative sign (-0.182), suggesting that DZs with more space produce fewer outputs, but it i s not significantly different from zero26. The number o f machines also i s not statistically significantly related to DZ output.

O f the case-mix variables, the proportion o f visits by children age 0-15 years i s statistically significant (p<0.05). The coefficient on this variable i s negative, indicating that a one percent increase in the share o f visits by children leads to a 0.03 percent decrease in total output, possibly indicating that visits by children require more time o f the DZ medical staff. The proportion o f visits by adult women i s weakly significant (p=0.098), and the sign i s positive, indicating that more visits by women slightly increase the productivity o f DZs.

The parameter “lambda” (sigma-u/ sigma-v) provides an estimate o f the share o f variance in output that i s attributable to productive inefficiency (Table A.1). If lambda i s estimated to be zero, inefficiency i s not a significant determinant o f variations in output. The estimate o f lambda in this sample i s 1.461, and a likelihood-ratio test based on a chi-squared distribution rejected the null hypothesis that this term i s equal to zero (p=0.019). This indicates that the error term should include the stochastic inefficiency component. The proportion o f total variance attributable to the inefficiency term i s equal to (sigma-u2/ sigma-u2+sigma-u2),

25 There are a large number o f zeros in the input variable for equipment (64 DZs, or 44 percent have a zero value). There are a number o f ways to deal with this issue, but none i s entirely satisfactory. A recommendation was made to substitute zero values with the smallest non-zero value divided by two. This i s considered to be a reasonable approach, because al l DZs have some basic equipment that was not measured in this survey. Therefore, basic equipment i s coded as 0.5 and each additional piece o f large equipment (x-ray or ultrasound machine) i s coded as an additional 1 unit of equipment. 26 However, the space coefficient i s significant if the regression excludes case-mix variables. A significant “space” variable i s relevant for the current health policy discussion to right-size health facilities, and reduce production costs.

39

which is estimated to be 0.682. This means that 68 percent o f total variance in output can be attributed to inefficiency. I t i s important to note that because some output data may be inflated, the production frontier may actually be lower than what i s estimated by the model, and thus be over-estimated.

86. None o f the factors hypothesized to affect efficiency are statistically significant (Table A.2). Thus, other factors that are unobservable, such as management style or staff motivation, may be more important determinants o f DZ efficiency.

Table A.l: Estimation of the Stochastic Frontier Production Function Variable Coefficient Z Prob > IzI

Dependent variable ln(tota1 # o f consultations at the DZ in 2007)

# o f observations 134

Wald chi2 23 1.84 Log likelihood -1 18.959

Prob > chi2 0.000 In (# o f physicians) 0.820 3.21 0.001*** In (# o f nurses) 0.264 0.91 0.361 In (# o f paramedical stafg 0.001 002 0.984 In(totaI space m2) -0.182 -1.48 0.140 In(tota1 # o f machines) 0.016 0.33 0.742 In(proportion o f visits by -0.03 1 -2.08 0.038" children 0-15 years) ln(proportion o f visits by 0.022 1.65 0.098* females 16- 1 5) Ln(proportion o f visits by -0.008 -0.47 0.639 adults 60 and over)

s igm a-v 0.445 sigma-u 0.65 1 s igm a2 0.622 lambda (sigma-u/ sigma-v) 1.461

chi-squared (1) = 4.30 Prob>chi-squared=O.O 19

Constant 10.13 11.22 0.000"'

* statistically significant at the 10% level; * * statistically significant at the 5% level; * * statistically significant at the 1% level.

Table A.2: Determinants of Inefficiency in DZ Production Variable Coefficient T Prob > It1

Dependent variable u (estimated inefficiency term) # of observations 134 R2 0.0058 F(5, 128) 0.15 Prob > F 0.980 Rural 0.018 0.35 0.728 Stand-alone -0.026 -0.43 0.670 Distance from hospital 0.0006 0.67 0.502 Catchment area size 6. 70e-06 0.08 0.93

Population density -9.62e-07 -0.07 0.946 Constant 0.504 6.41 o.ooo* * * (km2>

40

* statistically significant at the 10% level; ** statistically significant at the 5% level; * * statistically significant at the 1% level.

87. The stochastic frontier production function estimates can be used to calculate the level o f relative production efficiency for each DZ, which i s the ratio o f the total number o f consultations to the maximum possible output. The ranking o f DZs by production efficiency score i s presented in Figure A.1. An efficiency score o f 1 indicates that the maximum possible output has been achieved, so scores closer to 1 indicate more efficient DZs. Efficiency scores for DZs range from 0.136 to 0.866, with a mean o f 0.641 (median=0.640). Six DZs have efficiency scores below 0.40 raising concerns about their levels o f inefficiency, whereas 14 DZs report rather high levels o f efficiency with scores above 0.80.

Figure A.l: Ranking o f DZs by Production Efficiency Score

- 0.9 0 S 0.8 - 2 6 2 0.7 .g I: 0.6 E 0.5

0.4 9 0.3

S U I

.o .E 3 -g E 0.2

0.1 0 1 , ,

0 20 40 60 80 100 120 140 DZ

Results Cost Function

88. The DZ cost function i s estimated using SF analysis to assess the baseline cost efficiency o f DZs in producing their main outputs: consultations, laboratory tests, and diagnostic tests. The estimated coefficients reflect the marginal cost o f different DZ outputs (Table A.3). The coefficient on consultations can be interpreted as follows. A one percent increase in the number o f consultations leads to a 0.54 percent increase in DZ cost/expenditure, which i s significant at the one- percent level (p<O.OOl). The estimated coefficients on the number o f laboratory tests and diagnostic tests are very small and not statistically significantly different from zero. The average marginal cost o f consultations i s very low: 0.003 SD while the total average cost o f consultations i s 938 SD, which indicates that the vast majority o f DZ costs are fixed. None o f the case m i x variables was found to be statistically different from zero.

41

89. The estimate o f lambda (sigma-dsigma-v) i s 0.007 (Table A.3), and a likelihood- ratio test based on a chi-squared distribution failed to reject the null hypothesis that this te rm i s equal to zero (p=l.OOO). This indicates that the variance in costs observed across DZs i s not determined by differences in cost efficiency, and the error term should not include the stochastic inefficiency component. O f the factors hypothesized to affect cost efficiency, only population density o f the DZ catchment area i s statistically significant A higher population density in DZs catchment area i s associated with a higher level o f efficiency in DZs (Table A.4). The coefficient i s extremely small, -4.54e-", but it i s highly significant (p=O.OOO).

Table A.3: Estimation of the Stochastic Frontier Cost Function Variable Coefficient Z Prob > Iz/

Dependent variable # of observations Log likelihood Wald chi2

In(the total expenditure o f the DZ in 2007) 129

-2 17.44 20.02

Prob chi2 0.0012 In (# o f consultations in 0.544 4.35 o.ool*T 2007) In (# o f laboratory tests) 0.004 0.12 0.902 In (# o f diagnostic tests) 0.010 0.39 0.699 ln(proportion o f visits -0.002 -0.06 0.949 by children 0-1 5 years) In(proportion o f visits -0.009 -0.30 0.768 by.'females 16- 15) Ln(proportion o f visits 0.03 1 0.86 0.387 by adults 60 and over) Constant 12.02 2.10 0.036** s igm a-v 1.306 sigma-u 0,001 sigma2 1.705 lambda 0.007

chi-squared (1) = 0.00 Probkhi-squared= 1 .OOO

* statistically significant at the 10% level; * * statistically significant at the 5% level; * * * statistically significant at the 1% level.

Table A.4: Determinants o f DZ Cost Inefficiency

Dependent variable u (estimated inefficiency term) # o f observations 129 R2 0.145 F(5, 123) 4.17

0.0016 Prob > F Rural -5.99e'08 -1.33 0.186 Stand-alone -2.1 5e'08 -0.45 0.653 Distance from hospital -5.08e-lo -0.70 0.488

Variable Coefficient T Prob > It/

Population density -4.48e-' -3.76 0 .ooo"' Constant 0.0007 0.000"'

* statistically significant at the 10% level; * * statistically significant at the 5% level; ** statistically significant at the 1% level.

42

90. All DZs show a similarly high level o f cost efficiency. The mean o f the inefficiency error term i s 0.0009, and it does not vary significantly, with a range o f 0.000853 to 0.000855. The analysis cannot reject that an efficiency score o f 1 was achieved by al l o f the DZs, again reflecting that there was no variation in cost efficiency. The ranking o f DZs by cost efficiency score i s presented in Figure A.2.

Figure A.2: Ranking o f DZs by Cost Efficiency Score

1.000856 E

3 1.000855 E

.- 2

U

8 1.000852 0

0 20 40 60 80 100 120 DZ

4) Model Building Process

91. This section presents the alternative models shown in Table AS that were estimated in order to arrive at the final econometric specifications using an SF model presented in the previous Section 3. Given the weaknesses o f SF models and the lack o f consensus in the health economics literature on the validity o f the SF approach, this section compares the SF models to the estimation o f traditional productiodcost functions in the model-building process. In addition, alternative specifications o f functional form are tested under each approach. Based on the overall comparison o f alternative production model specifications, it i s concluded that the SF production model with a Cobb-Douglas functional form i s preferred. In addition, a plot o f the ranking o f DZs by the estimated production efficiency score shows little variation across alternative functional specifications. Similarly, a comparison o f the results o f the estimations o f the traditional cost functions and the SF cost models indicates that the SF model i s preferred.

92. Traditional cost and production functions are estimated with the variables that may affect efficiency included as explanatory “shifter” variables. These models are compared to the stochastic frontier approach, with the shifter variables included in the regression analysis o f the efficiency error term.

43

Functional Form

Translog Cobb-Douglas Granneman, Brown, and Paulv Cost Function

93. Different functional form specifications for the cost and production functions are examined. The starting point for the production function i s a translog specification, which i s a general and flexible functional form:

Approach Traditional Stochastic frontier model

Production cost Production cost X X X X X X X X -- X -- X

where,

lnY, lnL,i

Inspace, InEquipment, lnPropchild, InPropfem, InPropsenior, X"

The natural log o f the total number o f visits per year in DZ i27 The natural log o f the total number physician full-time equivalents in DZ i (input 1) The natural log of the total number nurse full-time equivalents in DZ i (input 2) The natural log o f the total number paramedical full-time equivalents in DZ i (input 3) The natural log o f the total space in square meters in DZ i (input 4) The natural log o f the total number o f machines in DZ i (input 5) The natural log o f the proportion o f visits by children 0-1 5 years in DZ i The natural log o f the proportion o f visits by females 16-59 years in DZ i The natural log o f the proportion of visits by adults 60 and above in DZ i A set o f variable that are expected to shift the production function (rural, standalone, catchment area, distance to the nearest hospital, population density) included directly in the traditional production function only

independent, identically distributed random error for DZ i = non-negative unobservable random variable associated with technical

inefficiency for DZi that i s assumed to follow a half-normal distribution (in the SF models only)

94. If the second-order input parameters are not found to be significantly different from zero, the functional form reduces to a Cobb-Douglas, which i s a special case o f the translog:

A Cobb-Douglas specification of the production function will be used, so the empirical estimate will be based on the natural logarithm o f the output (visits) and inputs (full-time equivalents for physicians, nurses, and paramedical staff).

27

44

95. The starting point for the cost function i s also the flexible translog specification:

lnc, =a; lnYCi +a, lnY, +a, lnY, +

where,

InPropchild, = InPropfem, = InPropsenior, =

X" - -

natural log o f the total annual expenditures o f D Z i natural log o f the annual total number o f consultations in D Z i (output 1) natural log o f the annual total number o f laboratory tests in D Z i (output

natural log o f the annual total number o f diagnostic services in D Z I

The natural log o f the proportion o f visits by children 0-1 5 years in D Z i The natural log o f the proportion o f visits by females 16-59 years in D Z i The natural log o f the proportion o f visits by adults 60 and above in D Z i A set o f variable that are expected to shift the production function (rural, standalone, catchment area, distance to the nearest hospital, population density) included directly in the traditional production function only independent, identically distributed random error for provider i

2)

(output 3)

non-negative unobservable random variable associated with cost inefficiency for provider i that i s assumed to follow a half-normal distribution (in the SF models only)

- -

96. Again, if the second-order input parameters are not found to be significantly different from zero, the functional form reduces to a Cobb-Douglas cost function, which i s a special case o f the translog:

97. A third, even less restrictive functional form for the cost function i s estimated that i s consistent with economic theory and accounts for the multi-product nature o f DZ output. The functional form, which i s based on the specification used by Granneman, Brown, and Pauly (1986) to estimate a hospital cost function, i s as follows :

45

lnc, =q lnY, +q lnY, +a, lnY, +

Comparison Group Translog vs. Cobb- Douglas (disaggregated staff variable)

Translog vs. Cobb- Douglas (aggregate staff variable)

Traditional Production Function Specifications

LR test statistic Prob > chi2 Conclusion LR chi2( 15) = 22.37 0.0985 Fail to reject the null hypothesis

that the higher order input terms are jointly equal to zero; Cobb- Douglas functional form i s preferred.

LRchi2(6) = 5.17 0.52 19 Fail to reject the null hypothesis that the higher order input terms are jointly equal to zero; Cobb- Douglas functional form i s preferred.

98. The results o f the estimation o f alternative traditional production functions are presented below in output tables (1) - (4). The estimation o f the traditional production function with a translog functional form (1) has a high R2 (0.698), but none o f the variables in the model were found to be statistically significant, including the variables that are hypothesized to affect efficiency. The coefficients on the staff variables (marginal product o f labor) have particularly wide confidence intervals. Because o f the small sample size, the additional higher order variables o f the translog functional form significantly reduce the degrees o f freedom. To gain efficiency, the model was estimated using an aggregate variable for the total number o f medical staff (physicians, nurses, and parmedical staff). The estimation results (2) s t i l l show that no variables in the model are statistically significant.

99. The traditional production function estimated with a Cobb-Douglas functional form (3) has a slightly lower R2 (0.643) due to fewer variables, but it yields a significant coefficient for the physician variable o f the appropriate sign. None o f the other variables in the estimate o f the traditional Cobb-Douglas production function are found to be statistically significant, other than the proportion o f pediatric consultations. Very similar results were obtained for the estimate o f the traditional Cobb-Douglas production function with the aggregate staff variable (4).

100. A likelihood ratio test o f the significance o f the higher order input terms in the translog model (Table A.6) failed to reject the null hypothesis that those terms are jointly equal to zero. Therefore, within the comparison o f specifications for the traditional production function, the Cobb-Douglas functional form with the disaggregated staff variable i s preferred.

46

(1) Traditional Production Function-Translog Functional Form

Source I ss df MS Number o f obs = 134 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - F ( 28 , 1 0 5 ) = 8 . 6 5

Residual I 3 9 . 9 6 0 1 5 7 3 1 0 5 . 3 8 0 5 7 2 9 2 7 R - squared = 0 . 6 9 7 6 Model I 9 2 . 1 7 9 4 2 6 5 28 3 . 2 9 2 1 2 2 3 7 Prob > F = 0 . 0 0 0 0

- - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Ad] R-squared = 0 . 6 1 6 9 T o t a l

l nconsu l t

lnphys lnnurse

lnpararned lnspace

Inequip2 input 11 input 2 2 input 3 3 input 4 4 input 5 5 input 1 2 input13 input14 input 1 5 input 2 3 input 2 4 input 2 5 input 3 4 input 3 5 input 4 5

d i s t-hosp r u r a l

standalone c a t c hsqkrn

popdensi ty lnconsped lncons f em l n c ons s en

cons -

1 3 2 . 1 3 9 5 8 4 133 . 9 9 3 5 3 0 7 0 5 Root MSE = . 6 1 6 9 1

6 . 1 5 1 9 2 5 - 7 . 5 3 9 4 5 8

, 1 7 9 7 5 7 2 1 , 9 7 1 4 3 1 - 1 , 0 9 3 5 7 - , 4 4 0 6 9 2 1 . 8 2 0 7 3 9 . 0 6 6 3 8 3 7 , 0 0 0 1 7 2 4 . l o 5 7 0 1 3 .1269393

- .0905494 - , 5 1 2 8 2 5 , 3 8 2 3 7 4 8

- . 1 6 8 8 8 0 1 - . 0 9 5 4 1 1

- . 5 5 0 9 5 0 7 . 1 2 3 2 9 8 2

, 2 3 9 9 6 4 5

. 1 1 2 5 4 0 4

. 0 7 0 5 8 8 7

.0000144 - . 0 2 9 6 3 0 1

. 0 1 5 4 7 4 9

. 0 0 2 5 9 5 9 8 . 7 0 3 3 8 4

- . 0191572

- . 0 0 0 2 2 6

6 . 5 3 e - 0 6

5 . 4 3 7 0 2 6 6 . 7 8 3 7 6 5 , 6 9 1 3 7 9 6 2 . 1 5 0 5 4 6 1 . 0 9 2 6 1 6 2 , 2 6 4 7 6 1 2 . 9 2 6 1 1 2 .0426845 .2759469 .2020655 2 , 3 3 5 1 7 2

, 1 6 8 2 4 1 , 6 6 9 2 2 0 6 , 3 1 5 6 8 0 3 . 2 0 7 0 1 2 5

.974195 .3802463 .0860194 .0358795 , 1 4 8 4 0 8 6 , 0 0 2 2 2 1 2 . 1 3 7 9 7 2 5 , 1 5 0 0 7 8 5 , 0 0 0 2 1 6 2

.000037 .0164589 .0154919 .0185615 1 2 . 3 5 5 0 1

1 . 1 3 -1.11

0 . 2 6 0 . 9 2

- 1 . 0 0 - 0 . 1 9

0 . 6 2 1 . 5 6 0 . 0 0 0 . 5 2 0 . 0 5

- 0 . 5 4 - 0 . 7 7

1 . 2 1 - 0 . 8 2 - 0 . 1 0 - 1 . 4 5

1 . 4 3 - 0 * 53

1 . 6 2 - 0 . 1 0

0 . 8 2 0 . 4 7 0 . 0 3 0 . 3 9

- 1 . 8 0 1 . 0 0 0 . 1 4 0 . 7 0

0 . 2 6 0 0 . 2 6 9 0 . 7 9 5 0 . 3 6 1 0 . 3 1 9 0 . 8 4 6 0 . 5 3 5 0 . 1 2 3 1 , 0 0 0 0 . 6 0 2 0 . 9 5 7 0 . 5 9 2 0 . 4 4 5 0 . 2 2 9 0 . 4 1 6 0 . 9 2 2 0 . 1 5 0 0 . 1 5 5 0 . 5 9 5 0 . 1 0 9 0 . 9 1 9 0 . 4 1 7 0 . 6 3 9 0 . 9 7 6 0 . 6 9 7 0 . 0 7 5 0 . 3 2 0 0 . 8 8 9 0 . 4 8 3

- 4 . 6 2 8 6 9 4 - 2 0 . 9 9 0 4 1 - 1 . 1 9 1 1 2 1 - 2 . 2 9 2 7 0 5 - 3 . 2 6 0 0 2 5 - 4 . 9 3 1 2 9 5 - 3 . 9 8 1 1 9 9 - . 0 1 8 2 5 1 8 - . 5469794 - , 2 9 4 9 5 7 3 - 4 . 5 0 3 2 7 6 - . 4 2 4 1 4 0 1 - 1 . 8 3 9 7 6 6 - . 2 4 3 5 6 0 8 - . 5 7 9 3 4 7 7

- 2 , 0 2 7 0 6 - 1 . 3 0 4 9 0 9 - , 0 4 7 2 6 2 2 - . 0902996 - . 0 5 4 3 0 2 4 - . 0046303 - . 1 6 1 0 3 3 6 - . 2269892 - . 0 0 0 4 2 2 2 - . 0000589 - . 0 6 2 2 6 5

- . 0 1 5 2 4 2 6 - , 0 3 4 2 0 8 1 - 1 5 . 7 9 4 3 2

1 6 . 9 3 2 5 4 5 . 9 1 1 4 9 5 1 . 5 5 0 6 3 5 6 . 2 3 5 5 6 7 1 . 0 7 2 8 8 4 4 . 0 4 9 9 1 1 7 , 6 2 2 6 7 8 .1510193 , 5 4 7 3 2 4 2 . 5 0 6 3 5 9 9 4 . 7 5 7 1 5 4 , 2 4 3 0 4 1 4 , 8 1 4 1 1 5 8 1 . 0 0 8 3 1

. 2 4 1 5 8 7 4 1 , 8 3 6 2 3 8 . 2 0 3 0 0 7 5 . 2 9 3 8 5 8 7 . 0 5 1 9 8 5 2 . 5 3 4 2 3 1 4 . 0 0 4 1 7 8 2 . 3 8 6 1 1 4 4 , 3 6 8 1 6 6 7 . 0 0 0 4 3 5 2 , 0 0 0 0 8 7 8 . 0 0 3 0 0 4 9 . 0 4 6 1 9 2 5

, 0 3 9 4 3 3 . 2 0 1 0 9

(2) Traditional Production Function-Translog Functional Form with Aggregate Staff Variable

Source 1 ss df MS Number o f obs = 134 F ( 1 7 , 1 1 6 ) = 1 2 . 5 6

Residual I 4 6 . 5 1 1 5 2 0 1 1 1 6 . 4 0 0 9 6 1 3 8 R-squared = 0 . 6 4 8 0

T o t a l I 1 3 2 . 1 3 9 5 8 4 133 . 9 9 3 5 3 0 7 0 5 Root MSE = . 6 3 3 2 2

- - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

Model I 8 5 . 6 2 8 0 6 3 7 1 7 5 . 0 3 6 9 4 4 9 3 Prob > F = 0 . 0 0 0 0

Ad] R-squared = 0 . 5 9 6 4 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

47

1 nequip2 input aa i npu tab i npu tac inputbb inpu tcc inputbc

d i s t-hosp r u r a l

standalone catchsqkm

popdensi ty lnconsped lncons f em lnconssen

cons -

- .6674284 .7367295

- .4293479 - . 2 1 9 5 4 9 1 - . 0 1 4 7 6 9 2

. 1 5 9 4 4 6 1 , 2 0 0 0 7 5 3

- , 0 0 1 0 1 7 4 - , 0 1 9 2 5 5 7

.1906912 , 0 0 0 0 2 1 3 2 . 4 8 e - 0 6

- , 0 3 4 5 6 6 1 . 0 2 7 9 6 2 9 - . 009268

- .4272314

.E455246

. 4 7 7 9 8 5 8

. 3 1 7 2 5 1 1

.1345653

. 2 5 0 7 7 3 4

. 2 0 3 7 0 4 7

. 1 4 0 2 5 1 8

. 0 0 2 2 2 2 1

. 1 2 9 1 6 2 9

. 1 4 4 5 6 5 1 ,0002173 . 0 0 0 0 3 5 4 . 0 1 6 6 2 7 1 . 0 1 5 4 2 3 8 . 0 1 8 3 2 1 1 7 . 3 3 3 2 0 1

- 0 . 7 9 1 . 5 4

- 1 . 3 5 - 1 . 6 3 - 0 . 0 6

0 . 7 8 1 . 4 3

- 0 . 4 6 - 0 . 1 5

1 . 3 2 0 . 1 0 0 . 0 7

- 2 . 0 8 1 . 8 1

- 0 . 5 1 - 0 . 0 6

0 . 4 3 2 0 . 1 2 6 0 . 1 7 9 0 . 1 0 5 0 . 9 5 3 0 . 4 3 5 0 . 1 5 6 0 . 6 4 8 0 . 8 8 2 0 . 1 9 0 0 . 9 2 2 0 . 9 4 4 0 . 0 4 0 0 . 0 7 2 0 . 6 1 4 0 . 9 5 4

- - - - - - - -

- 2 . 3 4 2 0 9 6 - . 2099815 - 1 . 0 5 7 7 0 4 - . 4 8 6 0 7 2 5 - . 5114576 - .2440167 - , 0 7 7 7 1 1 1 - . 0 0 5 4 1 8 6 - . 2 7 5 0 7 9 1 - . 0 9 5 6 3 8 1 - .0004092 - . 0 0 0 0 6 7 7 - . 0 6 7 4 9 8 1 - . 0025859 - * 0455553 - 1 4 . 9 5 1 5 6

1 , 0 0 7 2 4 1 . 6 8 3 4 4 1

,199008 ,0469744 .4819192

.562909 .4778617 .0033837 .2365678 .4770205 .0004517 .0000727

- . 0 0 1 6 3 4 1 ,0585117 .0270192 1 4 . 0 9 7 1

(3) Traditional Production Function-Cobb-Douglas Functional Form

Source I ss df MS Number o f obs = 134 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - F ( 1 3 , 1 2 0 ) = 1 6 . 6 0

Residual I 4 7 . 2 1 9 0 9 8 1 1 2 0 . 3 9 3 4 9 2 4 8 4 R - squared = 0 . 6 4 2 7

T o t a l I 1 3 2 . 1 3 9 5 8 4 133 .993530705 Root MSE = , 6 2 7 2 9

Model I 8 4 . 9 2 0 4 8 5 7 1 3 6 . 5 3 2 3 4 5 0 5 Prob > F = 0 . 0 0 0 0

- - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Ad] R-squared = 0 . 6 0 3 9

lnphys lnnurse

lnparamed lnspace

1 nequ i p 2 d i s t-hosp

r u r a l standalone c a t c hsqkm

popdensi ty lnconsped lncons f em lnconssen

cons -

.E783758 - 2 2 0 3 0 2 2 . 0 1 8 5 3 3 1

- . 1 9 8 0 6 7 9 , 0 1 4 5 7 9 3

- . 0 0 1 6 4 9 9 . 0 4 8 8 6 8 4 .0803838

- . 0001042 - . 000014

- . 0 3 5 0 5 1 8 . 0 1 9 1 4 6 5 . 0 0 1 5 3 1 6 9 . 7 7 0 7 5 7

. 2 9 4 4 8 3 5

. 3 2 5 6 8 1 9

. 0 3 2 0 0 7 5

. 1 3 6 0 1 0 7 .054487 . 0 0 2 0 9 2

.1322805 , 1 4 5 6 3 9 2 , 0 0 0 2 0 2 4 . 0 0 0 0 3 3 5 . 0 1 6 1 2 2 1

, 0 1 5 1 4 7 . 0 1 8 1 2 3 4 . 9 7 5 1 5 7 2

2 . 9 8 0 . 6 8 0 . 5 8

- 1 . 4 6 0 . 2 7

- 0 . 7 9 0 . 3 7 0 . 5 5

- 0 . 5 1 - 0 . 4 2 - 2 . 1 7

1 . 2 6 0 . 0 8

1 0 . 0 2

0 . 0 0 3 0 . 5 0 0 0 . 5 6 4 0 . 1 4 8 0 . 7 8 9 0 . 4 3 2 0 . 7 1 2 0 . 5 8 2 0 . 6 0 8 0 . 6 7 7 0 . 0 3 2 0 . 2 0 9 0 . 9 3 3 0 . 0 0 0

. 2 9 5 3 1 8 9 - . 4 2 4 5 2 5 3 - . 0448394 - , 4 6 7 3 5 9 6 - . 0 9 3 3 0 1 1 - . 0057919 - . 2 1 3 0 3 7 7 - , 2 0 7 9 7 1 6 - . 000505

- . 0 0 0 0 8 0 4 - . 0 6 6 9 7 2 4 - . 0108435 - . 0343515

7 . 8 4 0 0 1 3

1 . 4 6 1 4 3 3 .E651298 .0819057 .0712238 .1224598 , 0 0 2 4 9 2 2 .3107745 ,3687392 .0002965 .0000524

- . 0 0 3 1 3 1 1 . 0 4 9 1 3 6 5 .0374148 1 1 . 7 0 1 5

(4) Traditional Production Function-Cobb-Douglas Functional Form with Aggregate Staff Variable

Source I ss df MS Number o f obs = 134 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - F ( 11, 1 2 2 ) = 1 9 . 2 3

Residual 1 4 8 . 3 4 1 8 3 7 6 1 2 2 , 3 9 6 2 4 4 5 7 1 R- squared = 0 . 6 3 4 2

T o t a l 1 1 3 2 . 1 3 9 5 8 4 1 3 3 . 9 9 3 5 3 0 7 0 5 Root MSE = , 6 2 9 4 8

Model I 8 3 . 7 9 7 7 4 6 1 11 7 . 6 1 7 9 7 6 9 2 Prob z F = 0 . 0 0 0 0

- - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Ad] R-squared = 0 . 6 0 1 2

I n e q u i p 2 I d i s t - h o s p I

r u r a l I s t a n d a l o n e I

c a t c h s q k m I p o p d e n s i t y I

l n c o n s p e d I l n c o n s f e m I l n c o n s s e n I

c o n s I - . - - - - - - - - - - - -

Comparison Group 1 LR test statistic I Prob>chi2 Translog vs. Cobb- I LRchi2(15)= 24.81 I 0.0525

.0233374 - . 0 0 1 7 1 8 1 - ,0190353

.1381556 - . 0 0 0 0 7 5 7 - . 0 0 0 0 1 2 4 - . 0 3 7 5 8 4 1

.0235478 - . 0 0 5 1 7 9 7

8 . 2 1 6 4 3 2 , - - - - - - - - - -

Conclusion Narrowly fail to reject the null hypothesis that the higher order input terms are jointly equal to zero; Cobb-Douglas functional form i s preferred. Fail to reject the null hypothesis that the higher order input terms are jointly equal to zero; Cobb- Douglas functional form i s Dreferred.

.0542497

.0020989

.1264772

. 1 3 9 4 7 1 6

. 0 0 0 2 0 1 9

.0000334

.0160947

. 0 1 4 9 7 3 1

.0177206

.7581485 - - - - - - - - - - -

LRchi2(6) = 6.12

0 . 4 3 - 0 . 8 2 - 0 . 1 5

0 . 9 9 - 0 . 3 7 - 0 . 3 7 - 2 . 3 4

1 . 5 7 - 0 . 2 9 1 0 . 8 4

. - - - - - - - -

0.4102

0 . 6 6 8 0 . 4 1 5 0 . 8 8 1 0 . 3 2 4 0 . 7 0 9 0 . 7 1 1 0 , 0 2 1 0 , 1 1 8 0 . 7 7 1 0 . 0 0 0

. - - - - - - -

101.

102.

Stochastic Production Frontier Specifications

- .0840553 - . 0 0 5 8 7 3 1 - .2694096 - .1379423 - .0004754 - . 0000786 - , 0 6 9 4 4 5 1 - . 0 0 6 0 9 3 1 - , 0 4 0 2 5 9 5

6 . 7 1 5 6 0 1 . - - - - - - - - - - -

. 1 3 0 7 3 0 1

.0024368 .231339

. 4 1 4 2 5 3 5

. 0 0 0 3 2 4 1

. 0 0 0 0 5 3 8 - . 0 0 5 7 2 3 1

. 0 5 3 1 8 8 6

. 0 2 9 9 0 0 1 9 . 7 1 7 2 6 2

- _ - - - - _ _ _ - -

The results o f estimations o f alternative stochastic production frontier models are presented below in output tables (5) - (8). In the estimation o f the SF model with a translog functional form (5), as in the traditional production function with a translog functional form, none o f the variables in the model were found to be statistically significant, and the staff variables have particularly wide confidence intervals. Again, the translog model was estimated using an aggregate variable for the total number o f medical staff. The estimation results (6) s t i l l show that no input variables in the model are statistically significant, but the proportion o f pediatric consultations and the proportion o f adult female consultations become statistically significant (p<0.05). The proportion o f pediatric consultations has a negative effect on DZ output, and the proportion o f adult female consultations has a positive effect on output.

The SF production function estimated with a Cobb-Douglas functional form (7) yields a significant coefficient for the physician variable (p<O.OI), which i s o f the appropriate sign and similar in magnitude to the estimate o f the traditional production function with the Cobb-Douglas fwnctional form (0.878 vs. 0.820). None o f the other variables in the estimate o f the Cobb-Douglas SF production function are found to be statistically significant, other than the proportion o f pediatric consultations (p<0.05) and the proportion o f adult female consultations (p<O. 10). Very similar results were obtained for the estimate o f the Cobb-Douglas SF production function with the aggregate staff variable (4).

Douglas (disaggregated staff variable)

Translog vs. Cobb- Douglas (aggregate staff variable)

49

103, A likelihood ratio test o f the significance o f the higher order input terms in the translog SF production function model (Table A.7) failed to reject the null hypothesis that the higher order input terms are jointly equal to zero, although only narrowly (p=0.0525). Therefore, the Cobb-Douglas functional form with the disaggregated staff variable i s marginally preferred.

(5) Stochastic Production Frontier Model-Translog Functional Form

S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l N u m b e r o f obs = 1 3 4 Wald c h i 2 ( 2 3 ) = 3 0 9 . 0 8 - Log l i k e l i h o o d = - 1 0 6 . 5 0 0 9 3 P r o b > c h i 2 - 0 I 0 0 0 0

l n c o n s u l t

lnphys lnnurs e

1 np a r ame d lnspace

I n e q u i p 2 input11 i n p u t 2 2 i n p u t 3 3 input 4 4 input 5 5 input 1 2 input 1 3 i n p u t 1 4 input 1 5 input 2 3 input 2 4 input 2 5 input 3 4 input 3 5 i n p u t 4 5

lnconsped lnconsf em lncons sen

cons

_ _ _ _ _ _ _ _ - _ _ _ _

-

5 . 2 7 1 7 7 8 - 6 . 5 0 7 1 5 8

. 1 7 5 7 7 5 6 2 . 6 6 8 1 8 6

- 1 . 0 4 6 6 9 9 - . E 4 8 6 1 6 1 . 7 3 3 4 3 2 . 0 6 1 3 9 5 1

- . 0 7 9 1 1 0 1 . 0 9 0 3 2 4 6 . 3 1 2 5 7 2 7 - . 0 6 8 1 5 6

- . 3 3 4 0 5 9 1 . 3 5 3 4 4 8 1

- . 1 6 0 9 9 6 9 - , 2 4 5 5 0 2 2 - , 5 2 1 7 2 1 3

, 1 0 6 5 2 8 4 - , 0 1 1 1 1 3 2

. 2 3 3 0 6 9 6 - . 0 2 5 7 4 6 9

. 0 1 9 4 7 5 4 - . 0 1 0 0 1 2 4

5 . 6 2 2 7 6 9

4 . 3 7 9 2 1 9 5 . 5 0 5 3 3 9 . 5 6 3 3 4 8 9 1 . 8 6 5 4 6 5 . E 9 4 1 7 2 3 1 . 8 5 7 4 9 1 2 . 5 9 3 2 4 2 . 0 3 5 5 4 3 9 . 2 3 7 1 3 6 1 , 1 6 7 0 0 7 6 1 . 9 7 8 1 5 2 . 1 4 8 2 5 1 5 . 5 4 5 6 5 6 3 . 2 6 3 0 1 5 2 , 1 8 1 1 9 1 5 , 8 0 2 2 6 8 1 , 3 1 5 4 7 3 8

. 0 6 8 8 4 1

. 0 2 9 6 2 3 . 1 1 9 2 0 0 7 . 0 1 4 0 7 0 8 . 0 1 2 4 8 8 3

. 0 1 5 1 9 6 1 0 , 3 9 2 7 3

1 . 2 0 - 1 . 1 8

0 . 3 1 1 . 4 3

- 1 . 1 7 - 0 . 4 6

0 . 6 7 1 . 7 3

- 0 . 3 3 0 . 5 4 0 . 1 6

- 0 . 4 6 - 0 . 6 1

1 . 3 4 - 0 . 8 9 - 0 . 3 1 - 1 . 6 5

1 . 5 5 - 0 . 3 8

1 . 9 6 - 1 . 8 3

1 . 5 6 - 0 . 6 6

0 . 5 4

0 . 2 2 9 0 . 2 3 7 0 . 7 5 5 0 . 1 5 3 0 . 2 4 2 0 . 6 4 8 0 . 5 0 4 0 . 0 8 4 0 . 7 3 9 0 . 5 8 9 0 . 8 7 4 0 . 6 4 6 0 . 5 4 0 0 . 1 7 9 0 . 3 7 4 0 . 7 6 0 0 . 0 9 8 0 . 1 2 2 0 . 7 0 8 0 . 0 5 1 0 . 0 6 7 0 . 1 1 9 0 . 5 1 0 0 . 5 8 8

- 3 . 3 1 1 3 3 2 - 1 7 . 2 9 7 4 2 - , 9 2 8 3 6 8

- . 9 8 8 0 5 7 6 - 2 . 7 9 9 2 4 4 - 4 . 4 8 9 2 3 1 - 3 . 3 4 9 2 2 9 - . 0 0 8 2 6 9 7 - . 5 4 3 8 8 8 4 - - 2 3 7 0 0 4 3 - 3 . 5 6 4 5 3 4 - . 3 5 8 7 2 3 6 - 1 . 4 0 3 5 2 6 - . 1 6 2 0 5 2 3 - . 5 1 6 1 2 5 7 - 1 . 8 1 7 9 1 9 - 1 . 1 4 0 0 3 9 - . 0 2 8 3 9 7 5 - . 0 6 9 1 7 3 1 - . 0 0 0 5 5 9 6 - . 0 5 3 3 2 5 2 - , 0 0 5 0 0 1 2 - . 0 3 9 7 9 6

- 1 4 . 7 4 6 6 1

1 3 . 8 5 4 8 9 4 . 2 8 3 1 0 8 1 . 2 7 9 9 1 9 6 . 3 2 4 4 2 9 . 7 0 5 8 4 6 7 2 . 7 9 1 9 9 9 6 . 8 1 6 0 9 3

, 1 3 1 0 6 , 3 8 5 6 6 8 2 . 4 1 7 6 5 3 6 4 . 1 8 9 6 8

. 2 2 2 4 1 1 5

. 7 3 5 4 0 7 6

. E 6 8 9 4 8 5 , 1 9 4 1 3 1 8 1 , 3 2 6 9 1 4

. 0 9 6 5 9 6 . 2 4 1 4 5 4 3 . 0 4 6 9 4 6 7 . 4 6 6 6 9 8 7 . 0 0 1 8 3 1 4

. 0 4 3 9 5 2 . 0 1 9 7 7 1 3 2 5 . 9 9 2 1 5

- - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

/ l n s i g 2 v I - 2 . 0 5 4 5 7 4 . 3 5 8 8 2 5 6 - 5 . 7 3 0 . 0 0 0 - 2 . 7 5 7 8 5 9 - 1 . 3 5 1 2 8 9 / l n s i g 2 u I - . 7 5 9 2 7 3 3 . 3 4 0 8 0 3 9 - 2 . 2 3 0 . 0 2 6 - 1 . 4 2 7 2 3 7 - . 0 9 1 3 1

s i g m a - v 1 . 3 5 7 9 7 6 8 . 0 6 4 2 2 5 6 . 2 5 1 8 4 8 . 5 0 8 8 2 8 4 s i g m a - u I . 6 8 4 1 0 9 9 . 1 1 6 5 7 3 7 . 4 8 9 8 6 8 5 . 9 5 5 3 7 1 5

s i g m a 2 I , 5 9 6 1 5 3 8 . 1 2 8 8 4 7 3 , 3 4 3 6 1 7 7 . E 4 8 6 8 9 9 lambda 1 1 . 9 1 1 0 4 5 . 1 6 9 9 8 5 7 1 . 5 7 7 8 7 9 2 . 2 4 4 2 1 1

L i ke l i hood- ra t i o t e s t o f s igma-u=O: c h i b a r 2 ( 0 1 ) = 6 . 8 4 P r o b > = c h i b a r 2 = 0 . 0 0 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

(6) Stochastic Production Frontier Model-Translog Functional Form with Aggregate Staff Variable

S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l Number o f obs = 1 3 4 Wald c h i 2 ( 1 2 ) = 2 3 2 . 9 2

- Log l i k e l i h o o d = - 1 1 7 . 5 5 2 3 6 P r o b > c h i 2 - 0 . 0 0 0 0

50

- - - - - - - - - - - - - + - - - - - - - - - - - - - - - - _ - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

l n a l l s t a f f 1 1 . 0 2 2 1 9 8 1 . 1 1 7 0 6 5 0 . 9 2 0 . 3 6 0 - 1 , 1 6 7 2 0 9 3 . 2 1 1 6 0 5 lnspace 1 2 . 3 4 9 6 2 3 1 . 5 8 5 1 9 1 1 . 4 8 0 . 1 3 8 - . 7 5 7 2 9 5 5 , 4 5 6 5 4

l n e q u i p 2 I - . 7 3 0 6 1 7 5 . 7 3 2 7 3 8 7 - 1 . 0 0 0 . 3 1 9 - 2 . 1 6 6 7 5 9 . 7 0 5 5 2 3 9 inputaa I , 6 6 2 9 3 3 1 , 4 1 5 4 8 1 5 1 . 6 0 0 . 1 1 1 - . 1 5 1 3 9 5 8 1 . 4 7 7 2 6 2 inputab I - . 3 6 6 8 7 1 3 . 2 8 1 0 5 2 9 - 1 . 3 1 0 . 1 9 2 - . 9 1 7 7 2 4 9 . 1 8 3 9 8 2 2 i npu tac I - . 1 9 7 3 4 4 . 1 1 7 1 9 3 1 - 1 . 6 8 0 . 0 9 2 - . 4 2 7 0 3 8 3 . 0 3 2 3 5 0 3 inputbb I - . l o 4 8 2 1 3 . 2 3 1 5 5 6 5 - 0 . 4 5 0 . 6 5 1 - . 5 5 8 6 6 3 8 . 3 4 9 0 2 1 2 i n p u t c c I , 1 6 3 1 5 6 7 . 1 7 8 1 6 0 5 0 . 9 2 0 . 3 6 0 - . 1 8 6 0 3 1 3 . 5 1 2 3 4 4 8 inputbc I . 1 9 3 5 8 7 2 . 1 2 0 5 2 5 4 1 . 6 1 0 . 1 0 8 - . 0 4 2 6 3 8 2 . 4 2 9 8 1 2 7

lnconsped I - . 0 3 0 5 8 6 1 , 0 1 4 9 4 6 6 - 2 . 0 5 0 . 0 4 1 - . 0 5 9 8 8 1 - . 0 0 1 2 9 1 2 l n c o n s f e m I . 0 3 1 5 5 9 2 . 0 1 3 2 1 4 9 2 . 3 9 0 . 0 1 7 . 0 0 5 6 5 8 4 . 0 5 7 4 5 9 9 lnconssen I - . 0 1 7 8 9 0 7 , 0 1 6 1 6 8 7 -1.11 0 . 2 6 9 - . 0 4 9 5 8 0 7 , 0 1 3 7 9 9 4

- cons 1 - 1 . 0 4 5 2 6 3 6 . 5 0 9 2 1 4 - 0 . 1 6 0 . 8 7 2 - 1 3 . 8 0 3 0 9 1 1 . 7 1 2 5 6

/ l n s i g 2 v I - 1 . 7 3 9 9 9 9 . 3 0 3 7 1 9 9 - 5 . 7 3 0 . 0 0 0 - 2 . 3 3 5 2 7 9 - 1 . 1 4 4 7 1 9 / l n s i g 2 u I - . 7 4 7 6 0 7 1 . 3 6 5 4 9 7 2 - 2 . 0 5 0 . 0 4 1 - 1 . 4 6 3 9 6 8 - . 0 3 1 2 4 5 7

- - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

s i g m a - v I , 4 1 8 9 5 1 7 . 0 6 3 6 2 2 , 3 1 1 1 0 0 4 . 5 6 4 1 9 2 6 s i g m a - u 1 . 6 8 8 1 1 2 1 . 1 2 5 7 5 1 5 . 4 8 0 9 5 3 7 . 9 8 4 4 9 8 5

s i g m a 2 1 . 6 4 9 0 1 8 7 . 1 3 9 6 1 4 8 , 3 7 5 3 7 8 8 . 9 2 2 6 5 8 6 lambda 1 1 . 6 4 2 4 6 2 . 1 7 7 1 8 6 4 1 . 2 9 5 1 8 3 1 . 9 8 9 7 4 1

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . L i k e l i h o o d - r a t i o t e s t o f s igma-u=O: c h i b a r 2 ( 0 1 ) = 5 . 6 0 P r o b > = c h i b a r 2 = 0 . 0 0 9

(7) Stochastic Production Frontier Model-Cobb-Douglas Functional Form

S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l Number o f obs = 1 3 4 Wald c h i 2 ( 8 ) = 2 3 1 . 7 9

- Log l i k e l i h o o d = - 1 1 8 , 9 0 6 3 2 P r o b > c h i 2 - 0 . 0 0 0 0

- _ _ _ _ - _ _ _ _ _ - _ lnphys

lnnu rse 1 npa r ame d

lnspace 1 nequ i p 2

lnconsped lnconsf em lnconssen

cons -

+ - - - - - - - - - - - - - - - - - - - - - - -

I . E 1 9 7 2 4 . 2 5 5 7 3 3 2 1 . 2 6 4 4 9 8 6 . 2 8 9 6 0 4 5 I . 0 0 0 5 6 6 2 . 0 2 8 3 7 0 9

. 1 8 2 3 0 2 9 , 1 2 3 5 1 6 8

. 0 1 6 1 6 4 5 , 0 4 9 0 8 0 4

. 0 3 0 9 5 1 4 . 0 1 4 9 1 5 2 1 - . 0 2 1 6 1 5 9 . 0 1 3 0 6 9 8 1 - . 0 0 7 6 2 0 3 . 0 1 6 2 6 0 4 I 1 0 . 1 2 7 9 7 . 9 0 2 3 7 6 9

I -

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 . 2 1 0 . 0 0 1 . 3 1 8 4 9 6 1 0 . 9 1 0 , 3 6 1 - . 3 0 3 1 1 5 8 0 . 0 2 0 . 9 8 4 - . 0 5 5 0 3 9 8

- 1 . 4 8 0 . 1 4 0 - , 4 2 4 3 9 1 3 0 . 3 3 0 . 7 4 2 - . 0 8 0 0 3 1 4

- 2 . 0 8 0 . 0 3 8 - , 0 6 0 1 8 4 6 1 . 6 5 0 . 0 9 8 - , 0 0 4 0 0 0 4

- 0 . 4 7 0 . 6 3 9 - . 0 3 9 4 9 1 1 . 2 2 0 . 0 0 0 8 . 3 5 9 3 4 8

- - - - - _ - - - - 1 . 3 2 0 9 5 2

. E 3 2 1 1 3 , 0 5 6 1 7 2 2 . 0 5 9 7 8 5 5 . 1 1 2 3 6 0 4

- . 0 0 1 7 1 8 1 . 0 4 7 2 3 2 3 . 0 2 4 2 4 9 4

1 1 . 8 9 6 6

/ l n s i g 2 v I - 1 . 6 1 7 7 , 2 7 6 5 1 8 7 - 5 . 8 5 0 . 0 0 0 - 2 . 1 5 9 6 6 7 - 1 . 0 7 5 7 3 4 / l n s i g 2 u 1 - . E 5 9 3 2 7 5 . 3 9 9 0 8 0 6 - 2 . 1 5 0 . 0 3 1 - 1 . 6 4 1 5 1 1 - . 0 7 7 1 4 4

- - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

s i g m a - v I . 4 4 5 3 6 9 8 .Of515765 . 3 3 9 6 5 2 1 . 5 8 3 9 9 2 6 s i g m a - u I , 6 5 0 7 2 7 9 . 1 2 9 8 4 6 4 . 4 4 0 0 9 9 . 9 6 2 1 6 2 4

s i g m a 2 1 . 6 2 1 8 0 1 , 1 3 5 4 7 6 6 . 3 5 6 2 7 1 7 . E 8 7 3 3 0 4 lambda 1 1 . 4 6 1 0 9 5 , 1 7 9 0 2 1 2 1 . 1 1 0 2 2 1 . 8 1 1 9 7 1

L i k e l i h o o d - r a t i o t e s t o f s i g m a _ u = O : c h i b a r 2 ( 0 1 ) = 4 . 3 0 P r o b > = c h i b a r 2 = 0 . 0 1 9

(8) Stochastic Production Frontier Model-Cobb-Douglas Functional Form with Aggregate Staff Variable

S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l Number o f obs = 1 3 5 Wald c h i 2 ( 5 ) = 1 9 9 . 3 1


51

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . l n c o n s u l t I C o e f . S t d . E r r . 2 P > l Z I [ 9 5 % C o n f , I n t e r v a l ]

- - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

l n a l l s t a f f I . 9 5 6 6 8 3 . 0 7 4 0 8 8 7 1 2 . 9 1 0 . 0 0 0 . E 1 1 4 7 1 7 1 . 1 0 1 8 9 4 I n e q u i p 2 I . 0 0 5 8 1 9 7 . 0 5 1 4 3 1 3 0 . 1 1 0 . 9 1 0 - . 0 9 4 9 8 3 8 . l o 6 6 2 3 2

lnconsped I - . 0 4 3 5 2 9 4 . 0 1 5 2 6 4 6 - 2 . 8 5 0 . 0 0 4 - . 0 7 3 4 4 7 4 - . 0 1 3 6 1 1 4 l n c o n s f e m I . 0 1 9 1 3 1 2 . 0 1 3 3 5 3 4 1 . 4 3 0 . 1 5 2 - . 0 0 7 0 4 1 1 . 0 4 5 3 0 3 4 lnconssen I . 0 0 0 3 1 3 . 0 1 5 7 0 9 8 0 . 0 2 0 . 9 8 4 - , 0 3 0 4 7 7 7 . 0 3 1 1 0 3 7

- cons I 8 . 1 5 5 6 9 2 . 3 9 6 6 9 7 8 2 0 . 5 6 0 , 0 0 0 7 . 3 7 8 1 7 8 8 . 9 3 3 2 0 5 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

/ l n s i g 2 v I - 1 . 4 1 0 4 9 2 . 2 6 2 3 7 9 1 - 5 . 3 8 0 . 0 0 0 - 1 . 9 2 4 7 4 6 - . E 9 6 2 3 8 4 / l n s i g 2 u I - . 9 4 1 5 8 8 . 4 7 3 8 1 4 8 - 1 . 9 9 0 . 0 4 7 - 1 . 8 7 0 2 4 8 - . 0 1 2 9 2 7 9

s i g m a - v I . 4 9 3 9 8 7 . 0 6 4 8 0 5 9 s i g m a - u 1 . 6 2 4 5 0 6 2 . 1 4 7 9 5 0 2

s i g m a 2 I . 6 3 4 0 3 1 2 , 1 4 4 6 3 5 1 lambda I 1 . 2 6 4 2 1 6 , 2 0 0 3 2 8

, 3 8 1 9 8 5 4 . 6 3 8 8 2 8 5 , 3 9 2 5 3 7 2 . 9 9 3 5 5 6 9 . 3 5 0 5 5 1 7 . 9 1 7 5 1 0 8 . E 7 1 5 8 0 1 1 . 6 5 6 8 5 2

L i k e l i h o o d - r a t i o t e s t o f s igma-u=O: c h i b a r 2 ( 0 1 ) = 2 . 8 6 P r o b > = c h i b a r 2 = 0 . 0 4 5

104. A comparison o f the results o f the estimations o f the traditional production functions and the SF production models indicates that the SF model i s preferred. First, both approaches yield similar estimates o f the marginal productivity o f the inputs, with only physicians showing a marginal productivity statistically significantly different from zero in both approaches. In addition, the magnitude o f the coefficient i s similar in both estimation approaches. Second, likelihood ratio tests rejected the null hypothesis o f zero share o f the variance attributable to inefficiency (sigma-u=O) in each specification o f the SF model. This indicates that the error t e r m should include the stochastic inefficiency component.

105. The SF production model also appears to be robust to alternative specifications with regard to the inefficiency estimates, although the translog specification gives slightly higher estimates o f the share o f inefficiency in total variance than the Cobb- Douglas specification. The mean share o f the total variance attributable to the inefficiency te rm (sigma-u2/ sigma-u2+sigma-u2) ranges from 0.61 5 in the Cobb- Douglas specification with the aggregate staff variable to 0.776 in the translog specification with the disaggregated staff variable (Table A.8).

Table A.8. Estimates of Inefficiency as a Share o f Tota l Variance Across

106. In addition, a plot o f the ranking o f DZs by the estimated production efficiency score shows l i t t le variation across alternative specifications (Figure A.3). The Translog specification does yield marginally more variation in efficiency levels, and slightly higher inefficiency levels among poor performers, but these differences do not have any practical significance.

52

107 Based on the overall comparison o f alternative production model specifications, i t i s concluded that the SF production model with a Cobb-Douglas functional form i s preferred, because (i) likelihood ratio tests rejected the null hypothesis o f zero share o f the variance attributable to inefficiency; and (ii) likelihood ratio tests failed to reject the null hypothesis that the input interaction terms in the translog specification are joint ly equal to zero. Furthermore, due to the small sample size and large number o f parameters, the translog specification was not able to detect a positive marginal product for physician labor, which was estimated by the Cobb- Douglas specification.

Figure A.3: Ranking of DZs by Production Efficiency Score across Different M o d e l Specifications

_ _ _ - ~ ~ _ - - - P)

v)

CL

3

X

- 0.9 B g 0.8

E 0.7

.- E 0.6

2 0.5

0.4 u-

$

.- 2 0.2

5 0.1

0.3 0

0

.-

0 20 40 60 80 100 120 140 160

DZ

108. The SF production frontier model with the Cobb-Douglas functional form and disaggregated staff variable was also estimated using the service point, rather than the DZ, as the unit o f analysis. All variables in the model were deflated by the number o f service points in the DZ. The estimation results (9) are very similar to the model using DZ as the unit o f analysis, with a slight reduction in the estimated marginal product o f labor (0.820 vs. 0.758). The ranking o f DZs by the estimated production efficiency score shows l i t t le variation using service point as the unit o f analysis (Figure A.4).

(9) Stochastic Production Frontier Model-Cobb-Douglas Functional Form with Service Point as the Unit o f Analysis

S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l N u m b e r o f obs = 1 3 4 Wald c h i 2 ( 8 ) = 2 9 4 . 1 2


53

_ - _ _ _ _ _ _ _ - _ - - lnphysprsvpt

l n n u r s s v p t l n p a r a s v p t

l n s p a c e s v p t l n e q u i p s v p t

l n c o n s p e d l n c o n s f em l n c o n s s e n

c o n s -

+ - - - - - _ - - - - - - - -

1 . 7 5 8 1 3 1 5 I . 3 6 8 0 2 7 1 I - . 0 1 2 6 4 8 2

. 1 5 0 9 0 4 5 I - . 0 2 7 2 4 8 I - . 0 3 1 5 7 6 9 I . 0 2 2 2 7 9 9 I - . 0 0 7 9 3 0 8 I 9 . 5 9 6 5 9 1

- - - - - - - - - , 2 4 4 0 2 2 7 . 2 6 3 4 6 0 7 , 0 2 3 6 0 7 3

, 1 1 7 2 3 8 . 0 4 7 9 2 1 8 . 0 1 4 9 8 1 6 . 0 1 3 0 3 4 2 . 0 1 6 2 6 4 2

. 6 3 2 8 4 1

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 . 1 1 0 . 0 0 2 . 2 7 9 8 5 5 9 1 . 4 0 0 . 1 6 2 - . 1 4 8 3 4 6 4

- 0 . 5 4 0 . 5 9 2 - . 0589177 - 1 . 2 9 0 . 1 9 8 - . 3 8 0 6 8 6 8

0 . 5 7 0 . 5 7 0 - . 0 6 6 6 7 7 1 - 2 . 1 1 0 . 0 3 5 - , 0 6 0 9 4 0 4

1 . 7 1 0 . 0 8 7 - . 0 0 3 2 6 6 7 - 0 . 4 9 0 , 6 2 6 - . 039808 1 5 . 1 6 0 . 0 0 0 8 . 3 5 6 2 4 5

_ _ _ _ _ _ _ - _ c

1 . 2 3 6 4 0 7 .8844005 .0336213 , 0 7 8 8 7 7 8

. 1 2 1 1 7 3

, 0 4 7 8 2 6 5 , 0 2 3 9 4 6 4 1 0 . 8 3 6 9 4

- . 0 0 2 2 1 3 5

/ l n s i g 2 v I - 1 . 6 0 8 2 3 7 . 2 7 5 2 0 3 - 5 . 8 4 0 . 0 0 0 - 2 . 1 4 7 6 2 5 - 1 . 0 6 8 8 4 9 / 1 n s i g 2 u 1 - . 8 6 0 6 4 0 8 . 4 0 0 4 2 7 7 - 2 . 1 5 0 . 0 3 2 - 1 . 6 4 5 4 6 5 - . 0 7 5 8 1 7

s igma-v I . 4 4 7 4 8 2 2 . 0 6 1 5 7 4 2 s igma-u I . 6 5 0 3 0 0 7 . 1 3 0 1 9 9 2

s i g m a 2 I . 6 2 3 1 3 1 3 ' 1 3 5 7 4 4 4 l a m b d a 1 1 . 4 5 3 2 4 4 . 1 7 9 3 2 3 1

.5860064 , 3 4 1 7 0 3 3 . 4 3 9 2 2 9 9 . 9 6 2 8 0 1 , 3 5 7 0 7 7 1 .8891856 1 . 1 0 1 7 7 7 1 . 8 0 4 7 1 1

L i k e l i h o o d - r a t i o t e s t o f sigma-u=O: c h i b a r 2 ( 0 1 ) = 4 . 2 5 P r o b > = c h i b a r 2 = 0 . 0 2 0

Figure A.4: Ranking o f DZs by Production Efficiency Score across Different Units o f Analysis (Service Point)

1 Cobb-DougiasServicePolni I I

~ I Cobb-DouglasSDZ 0

0 20 40 60 80 100 120 140 160

DZ

Traditional Cost Function Specifications

109. The results o f estimations o f alternative traditional cost functions are presented below in output tables (1 0) - (1 5). In the estimation o f the traditional cost function with a translog functional form (1 0), none o f the variables were found to be statistically significant, with the exception o f population density (p<O.Ol), which has a negative sign, indicating that higher population density reduces DZ costs. The

54

coefficients on the output variables (marginal cost o f output) are not significantly different from zero and have particularly wide confidence intervals.

LRchi2(9) = 13.84

1 10. The estimation o f the traditional cost function with the Granneman functional form (1 1) gives similar results, with no coefficients significantly different from zero, with the exception o f two o f the second order terms and population density (p<O.Ol), which has a negative sign.

that the higher order output terms are jointly equal to zero; Cobb- Douglas functional form i s preferred. Fail to reject the null hypothesis that the higher order input interaction terms are jointly equal to zero; Cobb-Douglas functional form i s preferred.

0.1281

1 1 1. The traditional cost function estimated with a Cobb-Douglas functional form (1 2) yields a significant coefficient for the marginal cost o f consultations (p<O.O l), but none o f the other variables are found to be significant, again with the exception of population density (p<O.O l), which continues to be negative. Likelihood ratio tests o f the significance o f the higher order output terms in the translog and Granneman models (Table A.9) failed to reject the null hypothesis that the higher order terms are jointly equal to zero. Therefore, within the comparison o f specifications for the traditional cost function, the Cobb-Douglas functional form i s preferred.

Table A.9. Likelihood Rat io Test Statistics for Alternative Specifications o f the Traditional Cost Function

Comparison Group I LR test statistic I Prob>chi2 I Conclusion Translog vs. Cobb- I LRchi2(6) = 7.65 I 0.2647 I Fail to reject the null hypothesis Douglas

Douglas

(10) Traditional Cost Function-Translog Functional Form

Source 1 ss df MS Number o f obs = 1 2 9 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - F ( 1 7 , 111) = 3 . 1 0

Model I 8 2 . 8 1 3 7 8 6 3 1 7 4 . 8 7 1 3 9 9 1 9 Prob > F = 0 . 0 0 0 2 Residual I 1 7 4 . 6 0 5 3 5 8 111 1 . 5 7 3 0 2 1 2 4 R- squared = 0 . 3 2 1 7

T o t a l I 2 5 7 . 4 1 9 1 4 4 128 2 . 0 1 1 0 8 7 0 6 Root MSE = 1 . 2 5 4 2 - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Ad] R-squared = 0 . 2 1 7 8

h e x p t o t I Coef. S t d . E r r . [ 95% Conf , I n t e r v a l ]

l nconsu l t I n l a b 3

lnd iag2 ou tpu t11 ou tpu t 2 2 ou tpu t 3 3 output12 ou tpu t 1 3 ou tpu t 2 3

lnconsped

- 2 . 9 9 9 4 0 1 - . 0 0 9 4 2 5 3

. 0 1 9 0 8 2 8

. 2 7 4 9 5 4 9

. 0 1 9 4 4 6 2

. 0 0 1 3 2 0 1 - . 0 0 2 6 7 9 2

. 0 0 6 6 1 1 - . 0091759 - . 0368855

1 . 9 4 0 7 6 2 . 9 6 8 9 2 8 8

. 4 5 3 2 7 1 . 1 6 7 5 2 0 1 .0161413 . 0 1 5 7 7 7 5 . 1 4 1 4 6 2 7 , 0 7 0 5 7 0 7 . 0 1 0 5 4 2 1 . 0 3 5 3 8 0 8

- 1 . 5 5 0 . 1 2 5 - 6 . 8 4 5 1 5 1 - 0 . 0 1 0 . 9 9 2 - 1 . 9 2 9 4 2 2

0 . 0 4 0 . 9 6 6 - .E79104 1 . 6 4 0 . 1 0 4 - . 0 5 6 9 9 7 5 1 . 2 0 0 , 2 3 1 - . 0 1 2 5 3 8 8 0 . 0 8 0 . 9 3 3 - . 029944

- 0 . 0 2 0 . 9 8 5 - . 2 8 2 9 9 7 0 . 0 9 0 . 9 2 6 - . 1 3 3 2 2 9 6

- 0 . 8 7 0 . 3 8 6 - .0300658 - 1 . 0 4 0 . 2 9 9 - , 1 0 6 9 9 4 9

.E46349 1 . 9 1 0 5 7 2 . 9 1 7 2 6 9 6 . 6 0 6 9 0 7 2 . 0 5 1 4 3 1 2 . 0 3 2 5 8 4 3 , 2 7 7 6 3 8 5 . 1 4 6 4 5 1 6 , 0 1 1 7 1 4 1 , 0 3 3 2 2 3 9

55

l ncons f em lnconssen

r u r a l standalone

c a t chsqkm dist-hosp

popdensi ty cons -

.0171054

.0284284 - . 3230459 - . 0568452

. 0 0 0 2 0 9 - . 0 0 3 0 7 1 4 - . 0002596

3 3 . 9 8 6 7 8

. 0 3 2 1 3 4 5

. 0 3 7 5 9 8 9 . 2 6 5 7 5 4

, 3 2 9 4 9 2 8 . 0 0 0 4 1 1 5 , 0 0 4 3 3 1 2 , 0 0 0 0 6 7 2 1 3 . 9 9 9 7 7

0 . 5 3 0 . 5 9 6 - , 0 4 6 5 7 1 1 , 0 8 0 7 8 2 . l o 2 9 3 3 1 0 . 7 6 0 . 4 5 1 - , 0 4 6 0 7 6 3

- 1 . 2 2 0 . 2 2 7 - , 8 4 9 6 5 5 2 .2035634 - 0 . 1 7 0 . 8 6 3 - . 7 0 9 7 5 7 3 . 5 9 6 0 6 6 8

0 . 5 1 0 . 6 1 3 - . 0 0 0 6 0 6 4 .0010244 . 0 0 5 5 1 1 2

2 . 4 3 0 . 0 1 7 6 . 2 4 5 2 9 3 6 1 . 7 2 8 2 6

- 0 . 7 1 0 . 4 8 0 - . 0 1 1 6 5 4 - 3 . 8 7 0 . 0 0 0 - . 0 0 0 3 9 2 7 - . 0 0 0 1 2 6 5

Traditional Cost Function-Granneman Functional Form ss df MS Number o f obs = 1 2 9

- - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - F ( 2 0 , 1 0 8 ) = 2 . 9 5

(1 1) Source I Model I 9 0 . 9 9 3 0 8 4 3 20 4 . 5 4 9 6 5 4 2 1 Prob > F = 0 . 0 0 0 2

Residual I 1 6 6 . 4 2 6 0 6 1 0 8 1 . 5 4 0 9 8 2 0 4 R - squared = 0 . 3 5 3 5

1 nexp t o t

l n c o n s u l t l n l a b 3

lnd iag2 ou tpu t aa outputbb ou tpu t cc

ou tpu t aaa outputbbb ou tpu t ccc

ou tpu t ab outputac outputbc

lnconsped lncons f em lncons sen

r u r a l standalone

c a t c hsqkm d i s t-hosp

popdensi ty cons -

- 1 5 . 7 3 6 6 1 , 6 4 4 7 6 2 1

1 . 2 6 4 5 0 5 - , 0 7 4 5 4 1

- , 0 1 1 4 2 4 7 - . 0314452

. 0 0 5 3 5 2 2

. 0 0 0 5 6 1 1 - . 0629922

.0145004 - . 0055937

- , 0 3 2 6 7 .0243423 . 0 2 0 9 3 6 1

- . 1 5 3 2 1 6 4

- . 3 2 5 0 2 1 6 - . 0686798

. 0 0 0 0 7 4 4 - . 0039054 - . 0 0 0 2 7 9 2

8 3 . 8 2 5 8 6

1 8 . 1 1 3 4 7 1 . 0 6 7 9 2 3 , 5 3 4 9 2 6 9 1 . 4 8 2 5 6 5 . 0 4 0 7 8 7 1 . 0 3 4 1 5 3 1 , 0 4 0 7 9 7 9 , 0 0 2 5 4 4 1 . 0 0 3 5 4 1 3

. 0 8 0 4 9 8 , 0 3 5 6 4 3

. 0 0 5 4 7 0 9

. 0 3 5 5 0 1 4 , 0 3 2 1 3 6 3 . 0 3 7 3 8 2 1

, 2 6 3 1 4 6 . 3 2 6 5 1 2 3 , 0 0 0 4 1 3 1 . 0 0 4 3 0 7 5 . 0 0 0 0 6 7 3 7 5 . 4 0 3 9

- 0 . 8 7 0 . 6 0

- 0 . 2 9 0 . 8 5

- 1 . 8 3 - 0 . 3 3 - 0 . 7 7

2 . 1 0 0 . 1 6

- 0 . 7 8 0 . 4 1

- 1 . 0 2 - 0 . 9 2

0 . 7 6 0 . 5 6

- 1 . 2 4 - 0 . 2 1

0 . 1 8 - 0 . 9 1 - 4 . 1 5 1.11

0 . 3 8 7 - 5 1 . 6 4 0 6 5 0 , 5 4 7 - 1 . 4 7 2 0 4 7 0 . 7 7 5 - 1 . 2 1 3 5 3 4 0 . 3 9 6 - 1 . 6 7 4 1 9 7 0 . 0 7 0 - . 1 5 5 3 8 8 1 0 , 7 3 9 - . 0 7 9 1 2 2 1 0 , 4 4 3 - . 1 1 2 3 1 3 6 0 . 0 3 8 . 0 0 0 3 0 9 3 0 . 8 7 4 - . 0 0 6 4 5 8 3 0 . 4 3 6 - . 2 2 2 5 5 3 2 0 . 6 8 5 - , 0 5 6 1 5 0 1 0 . 3 0 9 - . 016438 0 . 3 5 9 - . l o 3 0 3 9 9 0 . 4 5 0 - . 0 3 9 3 6 1 4 0 . 5 7 7 - . 0531617 0 . 2 1 9 - .E466227 0 . 8 3 4 - . 7158838 0 . 8 5 7 - . 0 0 0 7 4 4 4 0 . 3 6 7 - . 0 1 2 4 4 3 7 0 . 0 0 0 - . 0 0 0 4 1 2 5 0 . 2 6 9 - 6 5 . 6 3 7 7 6

2 0 . 1 6 7 4 3 2 . 7 6 1 5 7 1 . g o 7 1 0 1 6 4 . 2 0 3 2 0 7

. 0 0 6 3 0 6 . 0 5 6 2 7 2 8 .0494233

.010395 , 0 0 7 5 8 0 5 , 0 9 6 5 6 8 6 . 0 8 5 1 5 0 9 . 0 0 5 2 5 0 6

, 0 3 7 7 . 0 8 8 0 4 5 9 . 0 9 5 0 3 3 9 . 1 9 6 5 7 9 5 , 5 7 8 5 2 4 3 .0008933 . 0 0 4 6 3 2 8

- . 0 0 0 1 4 5 8 2 3 3 . 2 8 9 5

(12) Traditional Cost Function-Cobb-Douglas Functional Form

1 nc onsul t l n l a b 3

Ind ia92 lnconsped lncons f em lnconssen

. 4 9 1 4 7 . 1 3 0 9 3 1 5 . 0 1 7 3 4 6 2 . 0 3 3 1 6 6 7 . 0 2 5 9 0 6 4 . 0 2 6 1 8 5

. 0 1 7 4 2 7 9 . 0 3 0 7 0 0 8 . 0 1 6 6 1 3 . 0 3 5 3 6 7 6

- . 0368532 . 0 3 4 7 8 7 -

3 . 7 5 0 . 0 0 0 . 2 3 2 1 6 7 . 7 5 0 7 7 2 9 0 . 5 2 0 . 6 0 2 - . 0 4 8 3 3 8 8 . 0 8 3 0 3 1 2 0 . 9 9 0 . 3 2 5 - . 0 2 5 9 5 1 6 , 0 7 7 7 6 4 5 1 . 0 6 0 . 2 9 2 - . l o 5 7 4 7 , 0 3 2 0 4 0 6 0 . 5 7 0 . 5 7 1 - . 0 4 3 3 7 3 5 . 0 7 8 2 2 9 3 0 . 4 7 0 . 6 3 9 - . 0 5 3 4 3 0 7 , 0 6 6 6 5 6 7

56

r u r a l I - . 3 7 8 0 4 1 6 .2568745 - 1 . 4 7 0 . 1 4 4 - . 886768

c a t c h s q k m I . 0 0 0 2 6 3 5 .0003975 0 . 6 6 0 . 5 0 9 - . 0005238 standalone I - . 1 5 8 4 5 8 2 . 2 9 8 6 5 5 1 - 0 . 5 3 0 . 5 9 7 - . 7 4 9 9 2 8 9

d i s t - h o s p I - . 0 0 3 4 5 9 2 . 0 0 4 2 4 5 3 - 0 . 8 1 0 . 4 1 7 - . 0 1 1 8 6 6 7 p o p d e n s i t y I - . 0 0 0 2 4 6 6 . 0 0 0 0 6 5 4 - 3 . 7 7 0 . 0 0 0 - . 0 0 0 3 7 6 2

- cons I 1 2 . 7 6 6 2 6 1.714988 7 . 4 4 0 . 0 0 0 9 , 3 6 9 8 1 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . _

.1306848

.4330125 , 0 0 1 0 5 0 8 . 0 0 4 9 4 8 4

- . 0 0 0 1 1 7 1 1 6 , 1 6 2 7 1

- - - - - - - - -

Stochastic Cost Frontier Specifications

112. The results o f estimations o f alternative stochastic cost frontier models are presented below in output tables (1 3) - (1 5). In the estimation o f the SF model with a translog functional form (1 3) and the Granneman functional form (1 4), as in the traditional cost function with a translog functional form, none o f the output variables in the model were found to be statistically significant, and they have particularly wide confidence intervals.

1 13. The SF cost function estimated with a Cobb-Douglas functional form (5) yields a significant coefficient on the output variable for consultations (p<O.O l), which i s o f the appropriate sign and similar in magnitude to the estimate o f the traditional production function with the Cobb-Douglas functional form (0.49 1 vs. 0.544). None o f the other variables in the estimate o f the traditional Cobb-Douglas production function are found to be statistically significant.

1 14. Likelihood ratio tests o f the significance o f the higher order input terms in the translog and Granneman SF cost functions (Table A. 10) failed to reject the null hypothesis that the higher order output terms are jo int ly equal to zero. Therefore, within the comparison o f specifications for the SF cost function, the Cobb-Douglas functional form i s preferred.

Table A.10. Likelihood Rat io Test Statistics for Alternative Specifications o f the

Comparison Group Translog vs. Cobb- Douglas

Granneman vs. Cobb- Douglas

Stochastic Fro1 LR test statistic

LR chi2(6) = 7.80

LR chi2(9) = 11.92

:ier Cost Function

that the higher order output terms are jointly equal to zero; Cobb- Douglas functional form i s

I preferred. 0.21 80 1 Fail to reject the null hypothesis

that the higher order input interaction terms are jointly equal to zero; Cobb-Douglas functional

(13) Stochastic Cost Frontier Model-Translog Functional Form

S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l Number o f obs = 1 2 9 Wald c h i 2 ( 1 2 ) = 3 1 . 4 4

Log l i k e l i h o o d = - 2 1 3 . 5 3 9 4 9 Prob > c h i 2 - 0 . 0 0 1 7 -

57

1 nc onsu l t l n l a b 3

l n d i a g 2 output 11 output 2 2 output 3 3 output 1 2 o u t p u t 1 3 o u t p u t 2 3

lnconsped l n c o n s f em lnconssen

cons

/ l n s i g 2 v / l n s i g 2 u

- - - _ - - - - - - - _ - _ + -

- 3 , 0 9 9 3 4 9 , 3 8 9 1 1 9

. 2 1 7 9 6 6 6

. 3 2 4 7 1 5 5

. 0 1 4 6 8 9 6

. 0 0 1 0 6 3 8 - . 0 5 8 7 5 6 9 - . 0 2 1 6 1 8 4 - . 0 1 6 3 4 4 2 - . 0 0 5 4 3 1 1 - . 0 0 7 2 8 8 4

. 0 4 7 3 4 8 7 3 1 . 0 8 2 4 2

. 4 7 2 8 0 4 7 . - - - - - - - - - -

- 1 0 , 2 5 0 4 3

. _

1 . 9 1 3 7 1 6 , 9 4 1 4 1 7 4 . 4 5 0 9 7 4 6 . 1 6 7 3 1 4 2 . 0 1 5 8 4 0 1 . 0 1 4 5 5 0 6 . 1 3 7 0 6 8 5 . 0 7 0 1 4 4 7 . 0 0 9 9 7 5 9 , 0 3 4 3 1 2 5 . 0 3 1 2 5 4 6

, 0 3 7 2 6 5 1 3 . 5 4 7 8 5

, 1 2 4 6 9 5 9 8 4 6 , 8 3 1 5

- - _ _ - - _ - _ -

. 6 5 1 4 6 5 1 - 1 . 6 2 0 . 1 0 5 - 6 , 8 5 0 1 6 3 0 . 4 1 0 . 6 7 9 - 1 , 4 5 6 0 2 5 2 , 2 3 4 2 6 3 0 . 4 8 0 . 6 2 9 - , 6 6 5 9 2 7 4 1 . 1 0 1 8 6 1 1 . 9 4 0 . 0 5 2 - . 0 0 3 2 1 4 3 . t i 5 2 6 4 5 3 0 . 9 3 0 . 3 5 4 - . 0 1 6 3 5 6 5 . 0 4 5 7 3 5 7 0 . 0 7 0 . 9 4 2 - , 0 2 7 4 5 4 8 , 0 2 9 5 8 2 4

- 0 . 4 3 0 . 6 6 8 - . 3 2 7 4 0 6 3 , 2 0 9 8 9 2 5 - 0 . 3 1 0 . 7 5 8 - , 1 5 9 0 9 9 6 . 1 1 5 8 6 2 8

. 0 0 3 2 0 8 1 - 1 . 6 4 0 , 1 0 1 - . 0 3 5 8 9 6 5 - 0 . 1 6 0 . 8 7 4 - . 0 7 2 6 8 2 5 . 0 6 1 8 2 0 2 - 0 . 2 3 0 . 8 1 6 - . 0 6 8 5 4 6 2 , 0 5 3 9 6 9 4

1 . 2 7 0 . 2 0 4 - , 0 2 5 6 8 9 3 , 1 2 0 3 8 6 8 2 . 2 9 0 . 0 2 2 4 . 5 2 9 1 1 1 5 7 . 6 3 5 7 2

, 7 1 7 2 0 4 1 3 . 7 9 0 . 0 0 0 - 2 2 8 4 0 5 3 - 0 . 0 1 0 . 9 9 0 - 1 6 7 0 . 0 1 1 6 4 9 . 5 0 9

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

s i g m a - v 1 1 . 2 6 6 6 8 4 , 0 7 8 9 7 5 1 1 . 1 2 0 9 7 9 1 . 4 3 1 3 2 7

s i g m a 2 I 1 . 6 0 4 5 2 3 . 2 0 0 7 0 1 6 1 . 2 1 1 1 5 5 1 . 9 9 7 8 9 1 lambda I , 0 0 4 6 9 3 3 2 . 5 2 2 6 6 6 - 4 . 9 3 9 6 4 4 . 9 4 9 0 2 7

s i g m a - u I . 0 0 5 9 4 5 2 . 5 1 7 1 8 7 0

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . L i k e l i h o o d - r a t i o t e s t o f s igma-u=O: c h i b a r 2 ( 0 1 ) = 0 . 0 0 P r o b > = c h i b a r 2 = 1 . 0 0 0

(14) Stochastic Cost Frontier Model-Granneman Functional Form

S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l Number o f obs = 1 2 9 Wald c h i 2 ( 1 5 ) = 3 6 . 6 4


l n e x p t o t I C o e f . S t d . E r r . [ 9 5 % C o n f . I n t e r v a l ]

lnconsul t l n l a b 3

l n d i a g 2 outputaa outputbb output c c

outputaaa outputbbb output ccc outputab output ac outputbc

lnconsped l n c o n s f em lnconssen I

cons 1 / I n s i g 2 v I / I n s i g 2 u I

- + - - _ - _ _ _ - - - - - - - -

- 3 . 4 7 9 6 9 7 1 . 0 4 7 2 5 5 . 2 4 4 5 2 8 7 . 2 4 7 7 6 0 9

- . 0 7 3 8 5 5 2 . 0 0 4 6 2 2 9

- . 0 0 2 3 8 0 7 , 0 0 5 1 3 0 9

- . 0 0 1 0 9 4 5 - . 0 9 0 6 1 0 7 - , 0 0 3 6 9 7 5 - . 0 0 9 8 7 7 8

. 0 0 2 6 4 6 - . 0 0 0 6 5 2 1

. 0 4 0 1 9 3 7 3 0 . 8 3 4 3 6

. 4 4 0 9 1 6 4 - 1 0 . 2 7 6 0 3

- - - - - - - - -

1 7 . 9 4 2 1 4 1 . 0 2 7 4 9 2 , 5 2 6 4 0 9 6 1 . 4 6 9 3 2 8 . 0 4 0 5 4 3 2 . 0 3 3 9 4 0 8 , 0 4 0 3 9 6 4 , 0 0 2 5 1 4 8 , 0 0 3 5 2 8 4 , 0 7 7 1 8 9 2

, 0 3 5 1 2 2 , 0 0 5 0 9 4 8 , 0 3 4 2 6 9 5 , 0 3 1 1 2 5 8 , 0 3 6 8 7 5 5 7 4 . 4 2 5 0 5

. 1 2 4 7 0 3 3 8 5 7 . 1 0 9 9

- - _ - _ - _ _ - - -

- 0 . 1 9 1 . 0 2 0 . 4 6 0 . 1 7

- 1 . 8 2 0 . 1 4

- 0 . 0 6 2 . 0 4

- 0 . 3 1 - 1 . 1 7 - 0 . 1 1 - 1 . 9 4

0 . 0 8 - 0 . 0 2

1 . 0 9 0 . 4 1

3 . 5 4 - 0 . 0 1

- - - - - - - - - -

0 . 8 4 6 - 3 8 . 6 4 5 6 4 0 . 3 0 8 - . 9 6 6 5 9 3 0 , 6 4 2 - . 7 8 7 2 1 5 1 0 . 8 6 6 - 2 . 6 3 2 0 6 8 0 . 0 6 9 - . 1 5 3 3 1 8 4 0 . 8 9 2 - . 0 6 1 8 9 9 8 0 . 9 5 3 - . 0 8 1 5 5 6 1 0 . 0 4 1 , 0 0 0 2 0 1 9

- , 0 0 8 0 1 0 . 7 5 6 0 . 2 4 0 - , 2 4 1 8 9 8 7 0 . 9 1 6 - , 0 7 2 5 3 5 4 0 . 0 5 3 - . 0 1 9 8 6 3 4 0 . 9 3 8 - . 0 6 4 5 2 1 0 . 9 8 3 - . 0 6 1 6 5 7 5 0 . 2 7 6 - . 0 3 2 0 8 0 9 0 . 6 7 9 - 1 1 5 . 0 3 6 1

. . . . . . . . . . . . . . . . . . . . . . 0 . 0 0 0 . 1 9 6 5 0 2 4 0 . 9 9 0 - 1 6 9 0 . 1 8

3 1 , 6 8 6 2 5 3 . 0 6 1 1 0 2 1 , 2 7 6 2 7 3

3 . 1 2 7 5 9 . 0 0 5 6 0 8

. 0 7 1 1 4 5 7

. 0 7 6 7 9 4 7 , 0 1 0 0 5 9 8 , 0 0 5 8 2 1 1 , 0 6 0 6 7 7 3 . 0 6 5 1 4 0 3 , 0 0 0 1 0 7 7

. 0 6 9 8 1 3 . 0 6 0 3 5 3 2 . 1 1 2 4 6 8 2 1 7 6 , 7 0 4 8

, 6 8 5 3 3 0 3 1 6 6 9 . 6 2 8

- - - - - - - - -

s i g m a - v 1 1 . 2 4 6 6 4 8 . 0 7 7 7 3 0 5 1 . 1 0 3 2 4 1 . 4 0 8 6 9 7

s i g m a 2 I 1 . 5 5 4 1 6 5 . 1 9 4 4 3 6 6 1 . 1 7 3 0 7 6 1 . 9 3 5 2 5 4 lambda 1 . 0 0 4 7 0 8 1 2 . 5 2 0 7 8 5 - 4 . 9 3 5 9 4 1 4 . 9 4 5 3 5 7

s i g m a - u I . 0 0 5 8 6 9 3 2 . 5 1 5 3 3 4 0

_ _ _ _ - - - - _ _ _ _ _ _ - - - - - - _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - L i k e l i h o o d - r a t i o t e s t o f sigma-u=O: ch ibar2 ( 0 1 ) = 0 . 0 0 Prob>=chibar2 = 1 . 0 0 0

58

(1 5) Stochastic Cost Frontier Model-Cobb-Douglas Functional Form

S t o c . f r o n t i e r n o r m a l / h a l f - n o r m a l m o d e l Number o f obs = 1 2 9 Wald c h i 2 ( 6 ) = 2 2 . 0 2

Log l i k e l i h o o d = - 2 1 7 . 4 4 1 0 6 P r o b > c h i 2 - 0 . 0 0 1 2 -

l n e x p t o t I C o e f . S t d . E r r . z P > l Z I [ 9 5 % C o n f . I n t e r v a l ] - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

l n c o n s u l t I , 5 4 3 8 8 5 8 . 1 2 4 9 1 3 2 4 . 3 5 0 . 0 0 0 . 2 9 9 0 6 0 5 . 7 8 8 7 1 1 1 l n l a b 3 I . 0041014 . 0 3 3 4 0 2 3 0 . 1 2 0 . 9 0 2 - . 0 6 1 3 6 5 9 . 0 6 9 5 6 8 8

l n d i a g 2 I . 0 1 0 1 5 0 1 . 0 2 6 2 5 6 3 0 . 3 9 0 . 6 9 9 - , 0 4 1 3 1 1 3 . 0 6 1 6 1 1 6

l n c o n s f e m I - . 0 0 8 9 5 0 1 . 0 3 0 3 3 1 1 - 0 . 3 0 0 . 7 6 8 - . 0 6 8 3 9 8 , 0 5 0 4 9 7 8 l n c o n s s e n I . 0 3 0 9 5 5 9 , 0 3 5 8 2 1 2 0 . 8 6 0 . 3 8 7 - . 0392524 . l o l l 6 4 2

c o n s I 1 2 . 0 6 1 6 2 5 . 4 8 3 9 3 6 2 . 2 0 0 . 0 2 8 1 . 3 1 3 3 2 2 . 8 0 9 9 3

/ l n s i g 2 v 1 . 5 3 3 3 0 2 .1245433 4 . 2 8 0 . 0 0 0 . 2 8 9 2 0 1 7 .7774023

l n c o n s p e d I - . 0 0 2 2 1 4 9 . 0 3 4 7 4 1 5 - 0 . 0 6 0 . 9 4 9 - . 070307 . 0 6 5 8 7 7 2

- - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

/ l n s i g 2 u I - 1 3 . 9 1 2 8 1 3 8 3 9 . 6 1 - 0 . 0 0 0 . 9 9 9 - 2 7 1 3 9 . 0 6 2 7 1 1 1 . 2 3

s igma-v I 1 , 3 0 5 5 8 5 . 0 8 1 3 0 0 9 1 . 1 5 5 5 7 8 1 , 4 7 5 0 6 4

s i g m a 2 I 1 . 7 0 4 5 5 2 . 2 1 2 3 9 2 3 1 . 2 8 8 2 7 1 2 . 1 2 0 8 3 4 s igma-u I . 0009525 6 . 5 9 1 2 3 6 0

lambda I . 0007296 6 . 5 9 3 4 8 1 - 1 2 . 9 2 2 2 6 1 2 . 9 2 3 7 2

L i k e l i h o o d - r a t i o t e s t o f sigma-u=O: c h i b a r 2 ( 0 1 ) = 0 . 0 0 Prob>=chibar2 = 1 . 0 0 0

1 15. A comparison o f the results o f the estimations o f the traditional cost functions and the SF cost models indicates that the SF model i s preferred. Both approaches yield similar estimates o f the marginal cost o f output, with only consultations showing a marginal cost that i s statistically significantly different from zero in both approaches. In addition, the magnitude o f the coefficient i s similar in both estimation approaches. Although the likelihood ratio tests failed to reject the null hypothesis o f zero share o f the variance attributable to inefficiency (sigma-u=O) in each specification o f the SF model, which indicates that the error term does not include a stochastic inefficiency component, the SF production'function does statistically confirm an inefficiency component, and the contrast between the production and cost function estimations i s a key result o f the study.

1 16. As in the SF production model comparison, the choice o f functional form does not appear to affect the inefficiency estimates or the ranking o f DZs by efficiency score (Figure A.5). Although the Cobb-Douglas functional form yields the lowest efficiency scores, the difference i s less than 0.01 and not practically significant. The Cobb-Douglas specification is therefore preferred, because there i s sufficient power to detect a positive marginal cost o f consultations.

59

Figure AS: Ranking o f DZs by Cost Efficiency Score across Different M o d e l Specifications

1.0015 -~ 1.001

1.0005

1.005 1.0045

- ~ _ _ - - _ _ _ - _ _ - ,

-

, I , I 1 I

1.004 I 1.0035

1.003 1.0025

1.002

Translog Granneman

- ~ - ~ _ _ _ _ _ I----

60

ANNEX TABLE 1: LIST OF PHC CENTERS SURVEYED, JANUARY - DECEMBER 2007 # Name Municipality District

1 2 3 4 5 6 7 8 9

I O 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47

DZ Dr Milorad Vlajkovic DZ Dr Sima Milosevic DZ Milivoje Stojkovic DZ Mladenovac D Z Novi Beograd D Z Milutin Ivkovic DZ Rakovica DZ Savski Venac DZ Sopot DZ Stari Grad D Z Vozdovac DZ Zemun D Z Zvezdara D Z Djordje Kovacevic DZ Vracar DZ Bor DZ Kladovo D Z Negotin DZ Kucevo D Z Petrovac D Z Jovan Serbanovic D Z Veliko Gradiste D Z Zabari DZ Zagubica DZ Bojnik D Z Lebane D Z Leskovac D Z Medvedja D Z Vlasotince D Z Bac DZ Mladen StojanoviC D Z Backi Petrovac DZ Becej D Z Dr Djordje Bastic DZ Temerin D Z Titel D Z Veljko Vlahovic D Z Zabalj DZ Alibunar DZ Bela Crkva DZ Kovacica D Z Kovin DZ Opovo DZ Pancevo D Z 1.oktobar DZ Vrsac DZ Lajkovac

Barajevo Cukarica Grocka Mladenovac Novi Beograd Palilula Rakovica Savski Venac Sopot Stari Grad Vozdovac Zemun Zvezdara Lazarevac Vracar Bor Kladovo Negotin Kucevo Petrovac na Mlavi Pozarevac Veliko Gradiste Zabari Zagubica Bojnik Lebane Leskovac Medvedja Vlasotince Bac Backa Palanka Backi Petrovac Becej Srbobran Temerin Titel Vrbas Zabalj Alibunar Bela Crkva Kovacica Kovin opovo Pancevo Plandiste Vrsac Lajkovac

Beogradski Beogradski Beogradski Beogradski Beogradski Beogradski Beogradski Beogradski Beogradski Beogradski Beogradski Beogradski Beogradski Beogradski Beogradski Borski Borski Borski Branicevski Branicevski Branicevski Branicevski Branicevski Branicevski Jablanicki Jablanicki Jablanicki Jablanicki Jablanicki Juznobacki Juznobacki Juznobacki Juznobacki Juznobacki Juznobacki Juznobacki Juznobacki Juznobacki Juznobanatski Juznobanatski Juznobanatski Juznobanatski Juznobanatski Juznobanatski Juznobanatski Juznobanatski Kolubarski

61

48 49 50 5 1 52 5 3 54 5 5 56 5 1 58 59 60 61 62 63 64 65 66 67 68 69 I O 71 12 1 3 14 75 16 17 18 19 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 91

DZ Lj ig DZ Mionica DZ Obrenovac DZ Osecina DZ Ub DZ B DZ Bogatic DZ Dr Darinka Lukic DZ Krupanj DZ Loznica DZ Mali Zvornik DZ Sabac DZ Cacak DZ Lucani DZ Ivanjica DZ Aleksinac DZ Doljevac DZ Gadzin Han DZ Merosina DZ Nis DZ Razanj DZ Dr Ljubinko Djordjevic DZ Bosilegrad DZ Bujanovac DZ Presevo DZ Surdulica DZ Vladicin Han DZ Vranje DZ Dr Obren Pejic DZ Bela Palanka DZ Dimitrovgrad DZ Pirot DZ Vladimirci DZ Smederevo DZ Smederevska Palanka D Z Dr Milan Bane Djordjevic DZ Cuprija DZ Despotovac DZ Jagodina DZ Paracin D Z Rekovac DZ Svilajnac DZ Brus DZ Cicevac DZ Krusevac DZ Sava Stanojevic DZ Vlastimir Godic DZ A DZ Novi Pazar DZ Raska

L j ig Mionica Obrenovac Osecina Ub B Bogatic Koceljeva Krupanj Loznica Mali Zvornik Sabac Cacak Guca Ivanjica Aleksinac Doljevac Gadzin Han Mer o s i n a Nis Razanj Svrljig Bosilegrad Buj anovac Presevo Surdulica Vladicin Han Vranje Babusnica Bela Palanka Dimitrovgrad Pirot Vladimirci Smederevo Smederevska Palanka Velika Plana Cuprija Despotovac Jagodina Paracin Rekovac Svilajnac Brus Cicevac Krusevac Trstenik Varvarin A Novi Pazar Raska

Kolubarski Kolubarski Kolubarski Kolubarski Kolubarski Kolubarski Macvanski Macvanski Macvanski Macvanski Macvanski Macvanski Moravicki Moravicki Moravicki Nisavski N isavs k i Nisavski Nisavski Nisavski Nisavski Nisavski Pcinjski Pcinjski Pcinjski Pcinjski Pcinjski Pcinjski Pirotski Pirotski Pirotski Pirotski Podrinjsko-Kolubarski Podunavski Podunavski Podunavski Pomoravski Pomoravski Pomoravski Pomoravski Pomoravski Pomoravski Rasinski Rasinski Rasinski Rasinski Ras i n s k i Raski Raski Raski

62

98 99

100 101 102 103 104 105 106 107 108 109 1 I O 1 1 1 112 1 I 3 1 I 4 1 1 5 1 I 6 1 I 7 1 I 8 1 I 9 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 I 3 6 137 138 139 140 141 142 143 144 145 146

DZ Tutin DZ Nikola Dzamic DZ Dr Janos Hadzi DZ Kula DZ Dr Marton Sandor DZ Subotica DZ Ada DZ Coka DZ Kanjiza DZ Kikinda DZ Novi Knezevac DZ Senta DZ Novi Becej DZ Secanj DZ Srpska Crnja DZ Zitiste DZ Dr Bosko Vrebalov DZ Milorad Mika Pavlovic Indjija DZ Pecinci DZ Rurna DZ Sid DZ Sremska Mitrovica DZ Stara Pazova DZ Arandjelovac DZ Batocina DZ Dr Vojislav Dulic DZ Kragujevac DZ Lapovo DZ Raca DZ Sveti Djordje DZ Blace DZ Kursumlija DZ Prokuplje DZ Zitoradja DZ Sjenica DZ Boljevac DZ Knjazevac DZ Sokobanja DZ Zajecar DZ Apatin DZ Odzaci DZ Sornbor DZ A r i l j e DZ Evelina Haverfild DZ Cajetina DZ Nova Varos DZ Pozega DZ Priboj DZ Prijepolje

Tutin Vrnjacka Banja Backa Topola Kula Mali Idjos Subotica Ada Coka Kanjiza Kikinda Novi Knezevac Senta Novi Becej Secanj Srpska Crnja Zitiste Zrenjanin Indjija P e c i n c i Ruma Sid Sremska Mitrovica Stara Pazova Arandjelovac Batocina Knic Kragujevac Lapovo Raca Topola Blace Kursumlija Prokuplje Zitoradja Sjenica Boljevac Knjazevac Sokobanj a Zajecar Apatin Odzaci Sombor A r i l j e Bajina Basta Cajetina Nova Varos Pozega Priboj Prijepolje

Raski Raski Severnobacki Severnobacki Severnobacki Severnobacki Severnobanatski Severnobanatski Severnobanatski Severnobanatski Severnobanatski Severnobanatski Srednjebanatski Srednjebanatski Srednjebanatski Srednjebanatski Srednjebanatski Sremski Sremski Sremski Sremski Sremski Sremski Sumadijski Sumadijski Sumadijski Sumadijski Sumadijski Sumadij ski Sumadijski Toplicki Toplicki Toplicki Toplicki Uzicki Zajecarski Zajecarski Zajecarski Zaj ecarski Zapadnobacki Zapadnobacki Zapadnobacki Zlatiborski Zlatiborski Zlatiborski Zlatiborski Zlatiborski Zlatiborski Zlatiborski

147 DZ Uzice Uzice Zlatiborski

63

ANNEX TABLE 2: VARIABLES IN THE PHC FACILITY SURVEY, JANUARY - DECEMBER 2007

~

Category Output variables

Input variables

Policy variables

Other control variables

Variable # o f visits in different service categories: consultation preventive immunization dental visit home visit injection overnight stay physiotherapy diagnostic service laboratory test # of visits by different age groups: adults, age 60 and older women, age 16-59 men, age 16-59 children, age 0- 15 # visits by diabetics # of individuals enrolled in the DZ total expenditure o f the DZ total expenditure on specific inputs: personnel, other supplies, utilities, maintenance # of staff in each category (physicians, nurses, paramedical, administrative, technical) # of machines (x-ray, ultrasound, CT, other) # o f beds # of rooms (consultatiodprocedure, laboratory, pharmacy, public space) # of vehicles % o f revenue from capitation externally-driven staff cuts (available from human resources strategy) whether the DZ i s public or private whether DZ i s part o f a hospital radius of the catchment area distance to the nearest hospital population density ruraliurban % of revenue from the health insurance fund % o f revenue from patients % of revenue from other sources iemographic structure o f DZ enrolled population

64

ANNEX TABLE 3: STAFFING IN PRIMARY HEALTH CARE CENTERS, JANUARY - DECEMBER 2007

Staff Category

Mean (range) Al l DZs Rural Urban Stand-alone I n Health

Center Total number of staff

Per DZ I 239.4 I 167.1 I 312.7 I 224.9 I 276.9 (48- 1,276) (50 - 692) (48 - 1,276) (50 - 1,276) (48 - 810)

Per service point 35.8 30.5 41.2 24.5 65.0

Per 1,000 population 5.4 5.7 5.2 5.5 5.2 (3.5 - 729) (3.5 - 584) (5.9 - 729) (3.5 - 195) (6.6 - 729)

- (2.8-9.8) (3.5-8.1) (2.8-9.8 (2.8-9.8) (3.1 -8.1) Total # of physicians

PerDZ I 61.8 I 41.3 I 81.5 I 60.0 I 6 6 (1

Per 1,000 population

Total # of nurses Per DZ

Per service point


Total # of paramedical staff Per DZ

Per service point


Total # of administrative staff Per DZ

Per service point

Per population

Total # o f technical staff

I (0-348) I ( 11 - 198) I (0-348) I (0-348) I (0-218) Per service point I 8.5 I 7.8 I 9.1 I 6.3 I 13.8

(0 - 169) (0.58 - 169) (0 - 128) (0 - 48) (0 - 169) 1.3 1.4 1.3 1.4 1.2

(0 - 3.0) (0.73 - 2.1) (0 - 3.0) (0 - 2.6) (0 - 3.0)

118.0 85.8 148.9 112.1 132.2

16.6 15.3 17.9 12.1 27.7

2.7 2.9 2.5 2.7 2.5

(0 - 682) (27 - 374) (0 - 682) (0 - 682) (0 - 440)

(0 - 263) (1.7 - 263) (0 - 254) (0 - 95) (0 - 263)

(0 - 5.6) (1.5 - 4.8) (0 - 5.6) (0 - 5.6) (0 - 4.8)

3.5 1.9 5.0 3.1 4.2

0.43 0.29 0.57 0.27 0.82 (0.01 -44) (0.01 - 19) (0.01 -44) (0.01- 44) (0.01 - 19)

(0.0005 - (0,001 - 6) (0.0005 - (0.0006 - 5) (0.0005 - 10) 10) 10)

0.06 0.05 0.07 0.05 0.07 (0 - 0.34) (0 - 0.30) (0 - 0.34) (0 - 0.30) (0 - 0.25)

13.7 10.9 16.4 14.0 13.2

2.7 2.4 3.1 1.7 5.2

0.35 0.40 0.30 0.39 0.25

(0 - 83) (0 - 66) (0 - 83) (0 - 75) (0 - 83)

(0 - 83) (0 - 66) (0 - 83) (0 - 13) (0 - 83)

(0 - 1.2) (0 - 1.2) (0 - 0.77) (0 - 1.2) (0 - 0.76)


I (0-256) I (0-98) I (0-256) I (0- 175) I (0 - 256) Per service point I 6.6 I 4.7 1 8.3 I 3.9 I 13.0

(0 - 256) (0 - 80) (0 - 256) (0 - 37) (0 - 256) 0.88 0.98 0.78 0.92 0.78

(0 - 2.3) (0 - 1.7) (0 - 2.3) (0- 1.8) (0 - 2.3)

65

ANNEX TABLE 4: QUESTIONNAIRE FOR PRIMARY HEALTH CARE CENTERS

Question 101 102

103

104

Answer Definitions and Remarks DZNumber DZName Name of the DZ

Pre-assigned number - filled in by CESID)

Name of the Municipality where DZ is located (e& Savski Venac, Jagodina., ,)

Name of the town where DZ i s located

. Municipality Code Municipality Town

I05 Name of District where DZ i s located (e.g.

District Sevemobacki, Toplicki, etc., ,) Code District

"Rural" - oooulation living in catchment area of the DZ

1 I O 111

112

106

113

114

. . ., Location (rural or urban) as classified by DZ

1) rural 2) urban

i s mainly rural; "Urban" -population living in catchment area o f the DZ i s mainly urban

115

122 square meters 1 the patients I I All rooms in DZ whose main ouruose IS different than

123 124 125

. . mentioned above e.g. public space, waiting rooms,

Public space, waiting area, and other administrative offices, change rooms, utility rooms, room, sq. meters heating, maintenance, garage, kitchen if any, etc.. . Visit date filled by CeSID Name of interviewer filled by CeSID

126

Code of data entry person

Code Interviewer Person interviewed Name and position of the interviewed person

66

127 Code position of interviewed person Name of person who entered data I filled by CeSID

Revenue from different sources of payers, in Serbian Dinars, from January 1,2007 to December 31,2007

\Me: I f t o u recc i tc t l in-kind donations, pleaw cstiniatc and inwrt cash-,altic of in-hind donation\ such a 4 di'iigs

Quest1

20 1 202 203

2 04

205

206

207

208

209

210

211

212

220

Answer Definitions and Remarks Do the numbers on revenues given below re fer to the period from January 1,2007 to December 31,2007. If yes, please proceed to question 205. 1) yes 2) no Start date, DD.MM.YYYY End date, DD.MM.YYYY

Put in the date in the form dd.mm.yyyy. Put in the date in the form dd.mm.yyyy.

designed period. Only funds paid on behalf of MoF. Since MoF executes all payments for Government o f Serbia, payments on behalf of other Ministeries and institutions should be in "MoH' part for MoH, "Other government" for others, etc ... Total ammount of funds paid by MoH in designed period. Including infrastructure investments, vertical programs, prevention, vaccination.. . Excluding donations, projects and services on market (e.g. check-ups of MoH

Total ammount of funds paid by other Ministries in designed period. Excluding

Ministry of Finance

Ministry of Health employees)

Other Ministries I donations, projects and services on market I Total ammount o f funds paid by

MunicipaUCity Government in-designed period Including infrastructure investments, vertical programs, prevention Excluding donations, projects and services on market Total ammount o f fimds paid by other Government institutions in designed period (Distnct, Province) Including infrastructure investments Excluding donations and services

Total ammount o f funds paid directly by individual patients in designed period, including copayments and marketed and on- demand services Including institutions and companies paying for marketed and on-demand

Municipality/City

Other Government on market

etc ....). Revenues not related to provision of Other (lump sum) healthcare services.

~~ ~~~ ~~~ ~~ ~ ~~

rotal, Serbian Dinars Control Total, sum of 206-219 0 Error for Total 0 Control H IF Revenues, sum of 206 and 207 0

should add up to sum over answers 206 to 219

67

221

222

Expenditures, in Serbian Dinars, from January 1,2007 to December 31,2007

Error for HIF Revenues 0 Money owed to DZ on January 1,2007

Money owed to DZ on December 31,2007

Total amrnount of funds owed to DZ at the beginning and at the end o f the observed period.

Question

301 302 303 304

305

306

307

308

309

310

31 I

312

313

314

315

316

317

318 319

320 32 1

Answer

1) yes Do the numbers on expenditures given below refer to the oeriod from Januarv 1.2007 to December 31,2007.

Description and Remarks

Goods and services

-Pharmaceuticals /Drugs

Total amount o f funds spent on goods and services in designed period. Should add up 311 to 318 Total amount of funds spent for drugs/pharmaceuticals in designed period, including donations and projects Total amount o f funds spent for medical material in designed period, including

(cleaning material, medical and other equipment maintenance, phone, internet,

medical and non medical equipment (e.g. computers). Including donations and

donations and projects and including

-Medical material

68

I donations and projects

322 323

Staff, beds, equipments \ate: pleaw report rvcragc anniirl number for the period from Jaiiiirry 1,2007 to December 31, 2007. L g . if jo i i started the )ear a i th 10 doctors but at the ciid of the year you had 8, the ruerage i59 doctors

outstanding loans by DZ Total obligations January 1,2007 Total obligations December 31,2007

Question

-number of Laboratory room

-Pharmacy room

-Consultation and procedures rooms

-Public space and otber rooms

401

402

403

404

405

Laboratory room - all rooms in DZ whose main purpose i s linked to the laboratory activities Pharmacy room - all rooms in DZ whose main purpose i s linked to the pharmacy activities Consultation and procedures room - all rooms in DZ whose main purpose i s providing healthcare services to the patients All rooms in DZ whose main purpose i s different than mentioned above e g public space, waiting rooms, administrative offices, change rooms, utility rooms, heating, maintenance, garage, kitchen if any, etc

406

407

408

409

3

4

69

9

9 :I l l E .i

/ I l l I

- N m s;z s; m 0 v1 b 0 v1

Ten most frequent reasons for visits in DZ, total number per year

I'letise enter for the total of 12 months the most frequent reasons for visit (diagnoses)

Rank I i s the most frequriitly swn diagnosis (c.g. respiratory inliwion)

Rank 10 i s the least freqiierit ainong the ten diagnoses

-X-ray machine Number 410 1

2 3 4 r;

Question 60 1 602 603 604 605 606 607 608 609 Irin

Year of Year of Production Purchase Price

" I " , ." I 1 I I 611 Total number of top 10 diagnosis

9 10 11 12 13

Generate list of machines with following information for every machine: (1) year o f production; (2) year purchase; (3) price

73

411

3 4

-Ultrasound machines I Production I Purchase I Price 1 - I I I

5 6

8 9

I I I 10 11 12 13

18 19 20 ...

Generate list o f machines with following information for every machine: (1) year o f production; (2) year purchase; (3) price

25 26 *" I I I

74

412

1 1

Year of Year of -CT Production Purchase Price 1 2 3 4 5 6

12 13 14

413

15 1 16 I 1 3 I 1

-Vehicles Production Purchase Price 1 2

26 21 28

Generate list of machines with following information for every machine: (1) year of production; (2) year purchase; (3) price

19 20

Generate list of machines with following information for every machine: (I) year of production; (2) year purchase; (3) price

28 29 30

75

414

3 4

Year of Year of -Other Production Purchase Price 1

" I I I

9

Generate list of machines with following information for every machine: (1) year of production; (2) year purchase; (3) price

76

report no. 45111-yf serbia baseline survey on cost and efﬁ

Documents