essays in development economics and finance

Trang 1

_ ESSAYS IN DEVELOPMENT ECONOMICS AND FINANCE

A DISSERTATION

SUBMITTED TO THE DEPARTMENT OF ECONOMICS AND THE COMMITTEE ON GRADUATE STUDIES

OF STANFORD UNIVERSITY

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

Trang 2

UMI Number: 3292437

INFORMATION TO USERS

The quality of this reproduction is dependent upon the quality of the copy submitted Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction

In the unlikely event that the author did not send a complete manuscript

and there are missing pages, these will be noted Also, if unauthorized

copyright material had to be removed, a note will indicate the deletion

® UMI

UMI Microform 3292437

unauthorized copying under Title 17, United States Code ProQuest Information and Learning Company

300 North Zeeb Road

Trang 4

I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy

hpej Hehsyôn

(Aprajit Mahajan) Principal Adviser

(o Mo

(Ran Abramitzky)

hw “| aA) A

(Giacomo de Giorgi) J

Trang 5

This dissertation consists of four separate contributions to the fields of development economics

(Chapters 1 and 2) and finance (Chapters 3 and 4)

In Chapter 1, I analyze the unintended impact of health sector decentralization on a critical public health intervention, routine childhood immunization, using a policy experiment in one of the largest states of India I find that immunization rates fell due to this reform, and show evidence of public sector effort: reallocation towards clinical services The results imply that decentralization reforms should take the entire mandate of service provision into account

Chapter 2 is written in collaboration with Soohyung Lee and Azeem Shaikh Randomized field experiments are increasingly popular in the development literature However, program evaluations that rely on individual tests of multiple outcomes related to.a single treatment are particularly vulnerable to false positives, and may therefore overstate the effect of the program We propose an adaptation of multiple hypothesis testing procedures based on permutation testing methods to the randomized policy evaluation setting We illustrate the procedure by re-evaluating the impact of randomly-assigned political reservations for women in India

Chapter 3 asks how financial literacy affects decisionmaking This paper focuses attention on stock market participation or the lack thereof Using a broad-based assessment of financial literacy administered to a sample of older American respondents over the Internet under the RAND American Life Panel (ALP), I find a strong positive relationship between financial literacy and stock market participation, consistent with aversion to ambiguity From a policy perspective, financial education should be emphasized as Americans become increasingly responsible for their own long-term wealth and retirement well-being

Trang 6

Acknowledgement

I thank my principal advisor, Aprajit Mahajan, Ulrike Malmendier and Alessandro Tarozzi for advice and support, as well as the members of my committee, Ran Abramitzky, Giacomo di Giorgi and Timothy Bresnahan

I acknowledge with heartfelt appreciation the contribution of Nirmala and Mahesh Buch of Bhopal, Madhya Pradesh to Chapter 1 of this dissertation as well as the kind cooperation of Health Commissioner Dr Rajesh Rajora and RKS Officer Dr Abha Sahu in the design and dissemination of facility surveys

Research for chapters 3 and 4 of this dissertation was conducted while the author was a Sum- mer Associate at the RAND Corporation, Washington DC, and the author is grateful for its kind hospitality The author thanks Arie Kapteyn, Angela Hung and Jeff Dominitz for mentorship and

guidance during this time, Sandy Chien for direction regarding of the American Life Panel data,

Prakash Kannan for helpful discussions and Jaime Gassman, Raphael Godefroy, Victoria Hill, An- drew Radin and Guy Weichenberg for encouragement and support, Anna Maria Lusardi for careful reading and comments, and Erik Meijer for specific remarks on information aggregation The views expressed in this work do not necessarily represent the views of RAND

Special thanks to colleagues and friends Adam Cagliarini, Katja Kaufmann, Soohyung Lee, Maya Meidan, Sri Nagavarapu, Alejandro Ponce-Rodriguez and Azeem Shaikh as well as Pedro and Paula Miranda, Eleni Diamanti, Emeric Henry, Laura Lombardi, Enrique Seira and Fayaz Onn Also to the ladies of the Economics Department, particularly Patricia Luna, who is an inspiration A special thanks to Hao Wu, never forgotten

Trang 7

Preface iv Acknowledgement Vv 1 Decentralization Reforms and Immunization in Madhya Pradesh 1 1.1 Introduction 2.2 0 0 ee “ 1 1.2 Institutional Setting 2 ee 5

1.2.1 Immunizationin India 2 eee 5

1.2.2 The District Health System in Madhya Pradesh 2 0 ee 6 1.2.3 The Rogi Kalyan Samitt Reforms 2 2 ee kia 6

13 Data na a4 dán n.I adáa 8

1.3.1 Household Survey Đata: NEHS land 2 co 8

1.3.2 Administrative Records and Facility Surveys 0 0 ee ee 9

1.3.3 District-level characteristics © ee va 9

1.4 Empirical Framework 2 “_(a(aaAagaa 10

1.5 Main Findings 2 ee 11

1.5.1 Negative Program Effects On Immunization .-2 -204 11 1.6 Why Did The Program Hurt Immunization? 2 2 ee 15 1.6.1 Increased Quality of Clinical Services: Hospital Data 16 1.6.2 Increased Quality of Clinical Services: NFHS Individual Data .0 16 1.6.3 Negative Effects Related to Profitability 0 0.0.0.0 2 00040 17 1.6.4 Reduction in Village-Level Vaccination Outreach 2 2 ee 18

1.7 Discussion 6n Ha TA 19

1.8 Appendix 1A: Tables and Figures 0 ee 21

1.9 Appendix 1B: Construction of Variables 2 2.000 2 ee ee 36

Trang 8

1.10 Appendix 1C: Theoretical Framework 2 ee 2 Multiple Hypothesis Testing for Randomized Program Evaluations 2.1 2.2 2.3 2.4 Introduction ee Setup and Notation 2 0 ee Testing Procedures 2 6 0 c Q Q kh Q Q na cà v k cv cv Y k N k k Ko k va 2.3.1 Testing a Single Null Hypothesis 2 Q Q Q Q kg Quà vo 2.3.2 Testing Multiple Null Hypotheses 0.0.0.0 0200222200004 Empirical Applications: Gender and Public Goods In India

2.4.1 Program Description and Data 2 0 ee ee 2.4.2 Outcomes of Interest 2 ee 24.3 Results 2 eee Conclusion 60 -raaaa Appendix 2A: Tables and Figures 0 0 ee Appendix 2B: Replilcation ee 2.7.1 Polilical PartieipatOn cv cv cv vu v1 nà k kg va Na 2.7.2 Pradhan Characteristics 2.000.000 ee ee ee ee 3 Financial Literacy and Stock Market Participation 3.1 3.2 3.6 3.7 0n0nš3ia5 0 1 -— La TỤC

A Simple Model of Stock Market Participation 2 0.0.0.0 0.000 ee eae 3.2.1 Basic Portfolio Choice with CARA utility 2 0.0.0.0 20 0.2000 3.2.2 Illiteracy and Mistaken Beliefs 2 0 De ee ee ee Data: The RAND American Family Life Panel 2 022 ee ee 3.3.1 Sample Construction and Summary Statistics 2

3.3.2 Financial Literacy 2 ee

3.3.3 Risk Aversion 2 222 ee

Aggregate Measures of Financial Knowledge and Illiteracy .0 0 3.4.1 Principal Components Analysis (PCA) 0.0.00 0.000000 ee

Empirical Analysis 2 ee

3.5.1 Ordinary Least Squares / Probit Estimation .0 0 2008

Trang 9

3.8.2 Limitations and Future Research 0 eee 86

3.9 Appendix 3A: Tables and Figures ee ee ee 87

3.10 Appendix 3B: Measuring Risk Aversion inthe ALP .-. .2.-.04 101 3.10.1 Willingness to Gamble on Lifetime Income: Barsky, Juster, Kimball and

3100507221 HđaIA = 101

3.10.2 Multiple Price List: Holt and Laury(2002) 0 0 0 0004 102

3.10.3 Comparison of Methods 2 6 ees 103

4 The Value of Corporate Social Responsibility 105

4.1 Introduction 2 ee 105

4.1.1 Should We Expect CSR To Improve Firm Value? 2 0 0 20- 105

4.1.2 Existing Empirical Evidence 2 0 ee ee 106

4.2 Data and Sample Determination 2 0.0.00 eee ee ee 110 4217 EFPSE4Good Índexes ee vo 110 4.2.2 Are CSR Index Adds and Deletes Valid EZvent8? VỤ 110 42.3 Data ee ee ¬ 112 4.2.4 Estimation Method 2 0.0.0.0 22 ee ee ee 112 4.3 Results 0.00.0002 0.0.00.2.2.8 ee 114

4.3.1 Main Finding 20 2 een 114

4.3.2 Further Analysis Using Firm-Level Data 2 0.0.0 0 0.04 114

44 Discussion 2 ee 115

Trang 10

List of Tables 11 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 1.10 1.11 1.12 1.13 1.14 1.15 1.16 1.17 1.18 1.19 2.1 2.2 2.3 2.4 2.9 2.6 2.7 3.1

State-wise Immunization Rates, 1992 - 1998/9 ee 22

Main Somrce of Immunization in Madhya Pradesh, I998/9 23 Organization of the Rogi Kalyan Samiti 0 HQ kg kh V k kia 24 Typical Fee Schedule (Rs) 1996-2000: District Hospital, Bhopal .0 25 Average Budget Shares: Government and RKS, 2000 25 Summaxry Statistics: Household DataA Q Q Q Q Q Q HQ Q Q nu Q Q g v V k k kia 26

Program Effect on Total Vaccination 2 0 ee 27

Further Specifications 2 ee ee 28

Effect of Pre-Reform Characteristics on Delay in District Adoption 29 Counterfactual Program Effects On National Vs District Interventions 30 Means Comparison: Hospital Usage Statistics 2 0 ee ee ee hãi Effect of Early Adoption on Hospital Usage Statistics 2.2 2 Q Q Q Q Qua j1 Program Effect on Ôther Reproduective and Child Health Services 32 Effect of Barly Adoption on Government Hospital Visits(Children with Fever/Cough) 33 Program Efect on Total Vaccines By District/Commitbee Characterisiicsg 34 Effect of Early Adoption on Camps: Rural Vilages NEHS-2 35 Program implementation of Rogi Kalyan Samiti Q Q Q ee 39 Asset Indicators and Scoring Coefficients 0 0 ee ee ee es Al

Poverty Headcount Ratios for Madhya Pradesh (1993/4 and 1999/2000) Al

Trang 11

3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 4.1 1.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10 411

Respondent Answers for HRS/MS Comparable Financial Literacy Questions 90

Principal Components Analysis: Basic Literacy 0 2 ee 91 Principal Components Analysis: Stock-Related Investment Illiteracy 92

OLS Individual Estimates of Effects on Stock Market Participation 93

Probit Individual Marginal Effects on Stock Market Participation 94

OLS/ Probit Marginal Effects on Stock Market Participation 95

Effect of Including Additional Controls 2 0 ee 96 Testing For Valid Instruments 2 2 ee ee 97 Specification Tests: Endogenity of Stock-Market Related Illiteracy 98

Effect of Varying Measures of Risk Aversion 0.0 ee ee ee ee ee 98 Effect of Using A Planner 2 2 ee ee ee 99 Other Social and Behavioral Factors 2.2 ee ee 100 Comparison of CRRA parameters using 2 methods .00- 104 AII CSR-related Inclnsions/E/xclusions from E'PSE4Good UKI 118

AII COSR-related Inelusions/Exelusions from F'YSE4Good UK:H 119

Sample Size Per Review yGCÌle, cv Q Q Q HQ nạ kg ng kg kg k k kia 120 Cumulative Abnormal! Returns from Additions (4+3/-3) 2 .0 04 120 Cumulative Abnormal Returns from Deletions (+3/-3) 2.0 0 ee 120 Cumulative Abnormal Returns from Additions (+1/-1) 2 00 122 Cumulative Abnormal Returns from Deletions (¢1/-1) 122

Cumulative Abnormal Returns from Additions (4-5/-5) 2 .00 , 123

Cumulative Abnormal Returns from Deletions (+5/-5) 2 0 ees 128 Companies Affected by New Environmental and Human Rights Criteria 2 - 124

Trang 12

List of Figures

1.1 Map of Madhya Pradesh 2 ee ee sa 21

1.2 Districts of Madhya Pradesh 2 ee 21

Trang 13

Decentralization Reforms and

Immunization in Madhya Pradesh

1.1 Introduction

Does decentralization help or hurt public health services in developing countries? Decentralization as a means to improve public services has been rapidly gaining popularity over the last 25 years with more than 75 countries adopting some type of decentralization reform (Ahmad, Devarajan, Khemani, and Shah, 2005), often in response to the failure of service provision by centralized states (Bardhan, 2002) Some multilateral development agencies explicitly prioritize lending and technical assistance for decentralization-related programs (Bossert, 1998; Lipscomb and Mobarak, 2007; Ahmad, De- varajan, Khemani, and Shah, 2005) A growing number of advocates recommend decentralization or greater local-level autonomy when feasible and cost-effective health improving interventions fail to reach developing country populations (Victora, Hanson, Bryce, and Vaughan, 2004)

Trang 14

CHAPTER 1 DECENTRALIZATION AND IMMUNIZATION 2

Alternatively, elite groups may channel resources towards their preferred services (Bardhan and Mookherjee, 2006; Chaudhary, 2007) Thirdly, on the demand side, household takeup may be positively or negatively affected by new fiscal arrangements such as user-fee or tax regimes({Bardhan and Mookherjee, 2006)

In this paper, I analyze the impact of health sector decentralization on a critical public health intervention, routine childhood immunization, using a policy experiment in one of the largest states of India District hospitals in Madhya Pradesh gained the autonomy to raise and spend revenues from clinical services to supplement existing budget allocations, via new committees known collectively as the Hogi Kalyan Samiti The hospitals retained responsibility for coordinating district-level routine immunization, although funding and procurement of vaccination supplies remained centralized, and immunization remained free

I identify the effects of these reforms on individual takeup of immunization by exploiting the time-varying statewide implementation between two waves of a large household survey I capture individual-level exposure by matching district-level administrative data on reform starting dates to child-level birth and vaccination records from two waves of the Indian National Family Health Survey (NFHS-1 and NFHS-2) carried out before and after the reform Exposure is jointly-determined by birth date and district of birth and is hence plausibly exogenous to the individual I then estimate the impact of exposure on total vaccination in the pooled data, using a Tobit model with district, age and cohort fixed-effects

Total vaccination declined by 12-15% per year of exposure as a result of the reforms This reflects the joint outcome of changes in supply and changes in household take-up, but I argue that the former effect is dominant The unique institutional features of this reform therefore allow us to consider the impact of changes in provider incentives, rather than efficiency or price I show evidence suggesting that local decision-makers dedicated new resources towards improving curative services and reduced effort spent on vaccination, as an explanation for declining coverage In the case of full local autonomy with no constraints, therefore, we might anticipate an even larger deterioration in immunization coverage

Trang 15

low.Each child is entitled to a total of eight vaccinations (Polio 1-3, DPT 1-3, BCG and measles) but in 1998/1999, less than half of all children aged 12 to 23 months had received them, and 14% had received none at all

Secondly, the impact of health system decentralization and immunization is a key information gap in the public health literature (WHO, 2004) The relevant literature is relatively sparse Two recent program evaluations, Wagstaff and Yu (2007) and Kremer, Bloom, King, Bhushan, Clingingsmith, Loevinsohn, Hong, and Schwartz (2006), find opposing results.‘ On one hand,Kremer, Bloom, King, Bhushan, Clingingsmith, Loevinsohn, Hong, and Schwartz (2006) find that, in Cambodia,

contracting-out management at the district-level successfully improves immunization rates when

immunization is an explicitly targeted service indicator.On the other hand, in the absence of any

official guidance on immunization, Wagstaff and Yu (2007) find a fall of 4% in average village-level

immunization rates following increases in autonomy at health facilities in China Qualitative case- study evidence, meanwhile, points to a negative impact on the provision of vaccination services In Uganda, Akin, Hutchinson, and Strumpf (2005) find a declining share of district health budgets to

public goods activities (including vaccination) after decentralization In Zambia, Bossert, Chitah,

and Bowser (2003) find that national average immunization coverage (as measured by the third doses of DPT and polio) fell from 82% to 63% , particularly in smaller, poorer and more rural districts However, the paucity of data and small sample size often limit the validity of the latter.”

This paper contributes to the existing literature in four ways Firstly, the analysis uses individual child-level data on immunization, while the majority of studies in this literature rely on data aggregated at the level of adminstrative units’ Without disaggregated data, the analysis is necessarily limited, particularly with respect to the control of potential of social factors that also impact immunziation (Pande and Yazbeck, 2003) Just as importantly, aggregate immunization rates are often highly unreliable This is particularly true in India, where government statistics are well- known to overstate vaccination rates Indeed, Balraj, Mukundan, Samuel, and John (1993) finds gross mismatches between household coverage surveys and administrative documents even at the

village-levelt

Secondly, decentralization often occurs against a backdrop of economic change, leading to identification problems in much of the existing literature In particular, studies that rely on pre- and

1The empirical literature on the relationship between health outcomes and decentralization outside the health sector itself has been also been mixed Galiani, Gertler, and Schargrodsky (2005) find that utility privatization improves water quality in Argentina and reduces child mortality from water-borne diseases On the other hand, Robalino, Picazo, and Voetberg (2001) shows that political decentralization is related to increases in child mortality, and Khaleghian (2004) finds that it is related to lower levels of childhood immunization ˆ

2In Bossert (1998), the available district-level data sources are partial, inconsistent and begin only after decentralization was well under way On the other hand Akin, Hutchinson, and Strumpf (2005) consider only districts with sufficient resources to have available workplans in 1995, leaving only 13 observations

Trang 16

post-comparisons such as Bossert, Chitah, and Bowser (2003) are likely to confound the effects of decentralization with contemporaneous events For example, the Argentine financial crisis led to decentralization reforms but also increases in poverty, a falloff in health insurance and the disrup- tion of medical supplies throughout the public hospital network (Ahmad, Devarajan, Khemani, and Shah, 2005) Similarly, in Georgia, extensive decentralization in the health system was accompanied by a resource shortage for vaccine surveillance, which contributed to the deterioration of immuniza-

tion (Djibuti, Rukhadze, Hotchkiss, Eisele, and Silvestre, 2007) In the case of Bossert, Chitah, and

Bowser (2003) , economic disruptions in Zambia and changes in donor funding may have affected the supply chain for vaccines Here, I explicitly control for confounding trends as well as district-level differences

Thirdly, from a policy perspective, these findings provide evidence that, contrary to the recom- mendations of Victora, Hanson, Bryce, and Vaughan (2004), it is not generally true that decentral- ized health systems are better at delivering comprehensive primary care services Furthermore, a large literature in developing country health has examined the impact of user fees and cost-sharing in government facilities from a demand-side point of view The majority of authors find that fees directly reduce the demand for health services in a variety of international settings (see Gertler, Locay, and Sanderson (1987); Osuga and Nordberg (1993); Collins, Quick, Musau, Kraushaar, and Hus-

sein (1996); Gilson (1997); Akashi, Yamada, Huot, Kanal, and Sugimoto (2004); Jacobs and Price

(2004); Mubyazi, Massaga, Kamugisha, Mubyazi, Magogo, Mdira, Gesase, and Sukwa (2006) among many others) This study provides evidence that user fees and cost-sharing may also have a negative indirect effect in a multitask setting, reducing the coverage of services that are not themselves fee-bearing

Finally, the reform itself is of particular interest and an evaluation is timely The Rogi Kalyan Samiti(henceforth RKS) is an attractive model for reform as it is easy to implement and compatible with scaling up in the presence of resource constraints, leading to much positive publicity from

local policymakers as well as some international organizations (see for example UNDP(2002) and Transparency International India(2005)) Kumar (2004) finds, using a case study of 9 hospitals,

that the main benefit of the RKS has been the improvement in hospital environments but does not formally evaluate the impact on services®.Nonetheless, proposals for replication in other states have been advanced even in the absence of formal impact evaluation

The rest of the paper proceeds as follows In Section 2, I provide background on Madhya Pradesh, the status of immunization over the period of study and the details of the decentralization reform Section 3 describes the data and estimation strategy Section 4 presents the main results and.checks for issues of selection and identification In Section 5, I discuss the proposed explanation Section 6 concludes The appendix contains further discussion of the variable construction methods as well

Trang 17

as a theoretical framework to guide the reduced-form empirical analysis

1.2 Institutional Setting

1.2.1 Immunization in India

India introduced universal routine immunization in 1978, with the aim of controlling six target diseases (diphtheria, pertussis, tetanus, poliomyelitis, typhoid and childhood tuberculosis) under its Expanded Programme of Immunization (EPI) Measles replaced typhoid in 1985, when the EPI was

renamed the Universal Immunization Programme (UIP) The UIP’s (unmet) goals were to achieve

85% coverage against these six diseases for all children by 1990 In 1983, the National Health Policy revised this target to universal immunization by 2000 In 1992, the UIP was integrated into the Child Survival and Safe Motherhood Programme (CSSM}, and subsequently in 1997 into the RCH Individual states receive their budget allocations from the national government under the RCH but manage their own distribution Table 1.1 shows the state-level immunization rates between 1992

(NFHS-1) and 1998 (NFHS-2), which demonstrate considerable heterogeneity

There is a disturbing divergence between Madhya Pradesh and other states.In 1992, only 25% of rural children and 46% of urban children under 36 months of age reported being fully vaccinated Madhya Pradesh is one of the few states in which the level of full immunization has not only

stalled but actually fallen to just over 16% in rural and 40% in urban areas (NFHS-1(1995), NEH§S-2(2001),Gaudin and Yazbeck (2006)) The declines in both rural and urban immunization

are comparable only to Jammu, where significant political turmoil occurred over this period, and Rajasthan, which is extremely poor

This fall takes place even in the presence of increasing levels of partial immunization.Results shown later in the paper suggest that these increases are driven mostly by an increase in polio vaccinations under the nationally-organized Pulse Polio Immunization (PPI) campaign The disparity between trends in partial and full immunization suggests an increasing failure of the district health system in Madhya Pradesh to deliver the remaining vaccines via routine immunization

In their analysis of this data, Gaudin and Yazbeck (2006) find no evidence of differences in under-

lying channels of diffusion between states, and conclude that the main driver of this heterogeneity is state-level policy efforts Unlike a small number of other states°, Madhya Pradesh implemented no policies explicitly targeting immunization (for better or worse) over this period that might explain the decline in immunization However, the district health system did undergo a unique major policy change, described in the following section

Trang 18

1.2.2 The District Health System in Madhya Pradesh

Madhya Pradesh is one of the largest, albeit poorest, states of India, located in Central India with a population of 60 million’ spread across 48 districts (see Figure 1.8) 75% of the population is rural

and 35% belong to a minority or scheduled caste/tribe (SC/ST)* The Planning Commission Report

of 1993 suggests that approximately 34% of the population is classified as living below the poverty line (BPL)

The public health system in Madhya Pradesh is organized around a network of district hospitals These hospitals provide clinical services to the public and coordinate lower-level health facilities in

rural areas” , a structure that is fairly typical of most developing country health systems (English,

Lanata, Ngugi, and Smith, 2006; WHO, 1992)

As the apex institution, district hospitals also manage district-level public health and implement primary health policies, including free universal immunization In Madhya Pradesh, specifically, procurement of vaccination supplies is funded and coordinated by the state, but delivery to the public is administered by the chief paedatricians at the local district hospital who are appointed chief district immunization officers Immunization services are provided onsite at clinics in the hospital and subsidiary rural health facilities, or through family health and welfare camps These are the primary sources of all vaccination in the state Table 1.2 shows that in 1998/9, approximately 85% of the population in both urban and rural sectors reported that they received most of their vaccinations from the district health system, either in a facility or at a camp A significant remaining minority

(7.5%) report receiving most vaccinations from the Pulse Polio campaign, while the remainder rely

on the private sector

1.2.3 The Rogi Kalyan Samiti Reforms

Prior to 1995, these district hospitals received fixed budget allocations from the state and provided all clinical services for free The civil surgeon (hospital administrator) had to refer decisions regarding purchasing and adminsitration to the state government!’ In 1994, following an outbreak of the plague in the neighboring city of Surat in Gujarat, a vermin-infested hospital in the district of

7By population, Madhya Pradesh ranks among the 25 largest countries in the world

8A Scheduled Caste or Scheduled Tribe is one officially identified in a table or “schedule” of ethnic minorities initially attached to the Indian Constitution, which awards specific rights and privileges to members of such tribes

®Rural areas are covered by three successively lower tiers of facilities - community health centers (CHCs), primary health centers (PHCs) and sub-health centers (SHCs) - for which the district hospital serves as the final referral The official guidelines for rural health facilities are as follows: the lowest tier or SHC supplies limited primary care One PHC, acts as a referral unit for 6 SHCs and provides one qualified medical officer per population of 30,000 (20,000 in hilly areas) as well as at least 6 beds for in patient services and referrals for 6 SHCs For every 4 PHCs, there should be CHC equipped with 4 medical specialists (including surgeon, physician, gynecologist, pediatrician), 30 beds, an operating theatre and a laboratory In Madhya Pradesh, only half the number of mandated rural facilities have actually been built, with a deficit of approximately 10,000 sub health centers, 1000 primary health centers and 200 community health centers (Bajpai, Dholakia, and Sachs, 2005; MTH, 2006)

Trang 19

Indore was shut down under pressure by the local media An emergency directive was invoked to allow a committee of medical officers, citizens and local elected officials to take temporary control over government assets and privately raise revenue on behalf of the hospital in order to fund a large- scale extermination operation Following the successful refurbishment of the hospital using private donations and fee revenue, the state government formalized a set of administrative guidelines for the formation of such committees in 1995 and directed all district-level hospitals to implement them In 1999, after the state had deemed the reform successful, this directive was extended to all public health facilities i.e the remaining civil hospitals, community health centers and primary health centers

A RKS at each hospital is formally constituted as a registered society, or non-governmental organization The societies have a large number of officially-mandated common objectives that fall into two broad categories: maintaining the hospital and overseeing the National Health Programmes, which include immunization(A full listing of the objectives is in Table 1.3) Under the policy, each hospital continued to receive pre-reform budget allocations from the state government!! To supplement this, each RKS may raise its own money via user-fees, leasing of hospital property, loans or donations The committee has full autonomy over the use of hospital assets and may spend its own income freely on capital or service staff, but have no authority over compensation or hiring of doctors, who remain employees of the state at fixed wages

The principal decisionmakers on the committee are the civil surgeon, the district Panchayat!” President and the chief district bureaucrat or Collector The society in fact consists of twin committees: an executive committee that makes decisions regarding the day-to-day affairs of the hospital on a bimonthly basis, and a large management committee that meets annually to plan the overall hospital budget and strategy The civil surgeon, the Panchayat President and the Collector sit on both committees In practice, the civil surgeon as secretary has significant control over the management

committee’s agenda!*.Further details of the membership roster are in Table 1.3

In 2000, the revenue of an RKS at the average hospital amounted to about 10% of the government

budget!*.On average, 75% of total hospital budgets were dedicated to doctors’ salaries, 14% to drugs,

4.5% for building maintenance and less than 1% for equipment purchase(see Table 1.5) The RKS

11 Ag a check on the implementation, hospital budgets from the state were requested from 1995 to 2005 However, only six hospitals were able to provide breakdowns of data from 1995, prior to the reform I find that in these six hospitals, while the rupee allocation for staff salaries decreased in two cases (one of which is likely to be a data entry error in the case of Dhar}, in general the budget allocations for staff, drugs, building maintenance and equipment remained similar or increased, The data do not allow more rigorous testing

12Panchayats are locally-elected councils formed at each level of intra-state administration - villages, block and district - that have the authority to allocate certain public goods See Chattopadhyay and Duflo (2004) for a description and history of the Panchayat system

13 A host of other stakeholders also have input via the management committee, but do not participate in the executive

Trang 20

incremental revenue is spent mostly on buildings and equipment Other expense includes unspent money put aside in savings accounts for future projects as well as non-salary-related staff incentives such as training programs

User fees accounted for 78% of revenue (based on the 2000 figures) The policy on user fees for

clinical services is as follows: All users are charged a flat fee for all outpatient visits with additional

fees imposed for further services Any patient below the poverty line is not subject to these additional

fees'*, but not from fees for consultation or the purchase of drugs In theory, the committee may set these fees at their own discretion!® In practice, fees are set: below the price for private provision of such services and there is little variation across hospitals At the time of writing, the outpatient consultation fee is almost universally either 2 or 5 rupees For the case of a simple blood sugar test, 26 hospitals reported a cost of 5-35 rupees with an average of 20, while the corresponding private sector fee for a comparable test was reported to be 40 rupees on average Only one complete retrospective fee schedule for the relevant period 1996-2000 was obtained, and it is represented in Table 1.4

As part of the RCH program, however, immunization and family planning clinics and camps were exempt from this reform and remained free of charge for all patients, regardless of status Doctors at the hospital continued to be responsible for onsite vaccination and organizing camps in addition to their clinical work In the next section, I estimate the impact of these reforms on immunizations received by children in Madhya Pradesh

1.3 Data

For this analysis, I use data from three sources Firstly, I pool two cross-sections of household survey data from the National Family Health Survey I merge this data with district-level administrative records and facility surveys, collected in Madhya Pradesh, and finally add district-level characteristics obtained from the 1991 census

1.3.1 Household Survey Data: NFHS 1 and 2

Data on child health and family demographics including vaccination status is obtained from the first two waves of the National Family Health Survey (NFHS), a widely-used nationally-representative

survey The first wave of the NFHS was carried out before the reform in 1992 (NFHS-1) and

repeated, with some modification in late 1998 and early 1999 (NFHS-2) The survey covers women of reproductive age (18-49), including a complete birth history and retrospective health histories

15It is not entirely clear that these selective income-based exemptions for clinical services were rigorously en- forced.There is no auditing mechanism at the hospital and users are left to self-identify In hospital visits, fees are clearly posted but exemption notices are not.Another concern is that patients who hold BPL cards may not actually be the poorest but this lies beyond the scope of this paper

Trang 21

for recent births Information on current health and socioeconomic status (including district of residence) for women is available in 1992 and 1998/9, while information on incidence of immunization

status as well as mortality and current morbidity is available for their children born between 1988- 1993 and 1996-1999 (the 4 years preceding NFHS-1 and the 3 years preceding NFHS-2)

1.3.2 Administrative Records and Facility Surveys

Starting dates and revenues for RKS at various district hospitals from 1995-1999 were compiled by the author in January 2007 from administrative sources located at the Department of Health

and Family Welfare, in Bhopal, Madhya Pradesh Administrative records on registration and initial

revenue collection dates were compiled from early bank account files for 38 of the 42 hospitals'” To supplement this data, retrospective surveys were circulated at the individual facility level in February 2007, and interviews were conducted with hospital staff and the Commissioner of Health in Bhopal 26 facilities returned completed surveys in June 2007, giving details of budgetary breakdowns, fees and equipment purchases as well as subjective perceptions of the reforms although the data is of varying quality

1.3.3 District-level characteristics

District characteristics were collected for years prior to the reform, contemporaneous with the first wave of household data Indicators of poverty, population density and demographic composition were compiled from Census 1991 data and the United Nations Human Development Report for

Madhya Pradesh (UNDP,1998)

Sample Characteristics

The final data consists of the 2 pooled cross-sections from the Madhya Pradesh samples of NFHS1 and NFHS-2, matched by district of current residence with the district-level information gathered above For purposes of our analysis I restrict the sample to living children aged 3 and under in both waves of the survey I drop 2 districts from the analysis where NFHS data is collected in only one wave of the survey I also remove children of mothers who are temporary visitors and recent migrants (for reasons to be discussed later) The final dataset used contains 3804 individuals with

nonmissing data, 1868 from NFHS-1 and 1936 from NFHS-2

Trang 22

Children aged 0-3 have received an average of 3-4 vaccines, as opposed to the mandated total of 8 The data illustrate that a large increase in polio vaccination, consistent with the introduction of Pulse-Polio between these waves, which is also reflected in an increase in the proportion of children reporting having received any vaccination from 61% to 81% However, the statistics also show a slight fall in the average number of non-polio vaccinations, i.e vaccinations conducted primarily though the district-health system, and a concurrent fall in the number of fully-immunized children

from 21% to 16%

As the NFHS does not include a measure of income but includes an asset roster, I compute a proxy for poverty by constructing asset indexes for each wave of the survey using principal components analysis I define a poverty threshold for the index based on sector-specific poverty headcount ratios

computed for 1993/4 and 1999/2000 by Deaton and Dreze (2002) and Deaton (2003), such that the

weighted percentage of household-level observations in each sector below the threshold corresponds to the poverty headcount ratio A more detailed discussion of the method and poverty measures in India is found in the appendix

1.4 Empirical Framework

The identification relies on the time-varying implementation of the reform as well as the latter wave of the survey The basic strategy is to estimate differences in status between children who are exposed to the reforms for different lengths of time, controlling for district, age and time effects

The basic estimating equation is

Yide = BTide + ¥eDe + VaDa + ¥mDm + Xi t+ Gi (1.1)

Here, Yige is the outcome for a child 7 living in district d at born in cohort c Tjg- measures exposure to the program in years Total individual-level exposure is captured by matching the district-level administrative data on reform starting dates to child-level birth dates and interview dates and computing the length of time the child is alive under the program Exposure is jointly- determined by birthdate and location, and may be plausibly regarded as exogenous to the individual I include Dg, district-level dummies and D,, dummies for child birth cohort as defined by the

survey wave!®, To control for characteristics that are correlated with coverage, I include X;, a vector

of child and household demographics, including a dummy variable P; that takes the value of 1 if the household is a poor household, To most flexibly capture age-effects, I include a full set of dummies for child age in months, although the results are robust to various parametric specifications for the

19

child’s age’’ Conditional on the inclusion of these covariates, the identifying assumption is that

18In later specifications I test the validity of this parsimonious approach as opposed to using alternatives such as - birth year dummies

Trang 23

there is no trend that changes coverage occurring simultaneously with these reforms The coefficient @ then measures the average effect of a year of program exposure on the dependent variable

To find the average effect of the reform on immunization, I estimate Equation 1.1 using the main dependent variable of interest, total vaccinations The effects of censoring on dependent variables of interest may be significant, For binary outcomes, such as full-immunization, we use a probit specification for Equation 1.1 For censored variables, such as the total number of vaccines which is bounded below at 0 and bounded above at 8 respectively, we use a Tobit specification

Estimating program exposure is not straightforward given the data As the hospital committees often begin actual operations some time after they are authorized to do so, I have two alternative measures of total program exposure, which I refer to as de jure exposure (based on date of registra-

tion) and de facto exposure (based on date of revenue-collection) Details of the differences between

these measures are found in the appendix

Note that de facto exposure may well be endogenous, in that certain districts may find it easier to begin revenue collection due to factors that are correlated with later program implementation For example, the civil surgeon or collector in a particular district may be particularly able, or the population may be wealthier and hence more receptive to fees While this is a concern, the effect of time-invariant district-specific factors?° will be controlled for by the use of district-level fixed-effects

1.5 Main Findings

1.5.1 Negative Program Effects On Immunization

The main findings of the paper are presented in Table 1.7 In Column 1 and 2 I first estimate Equation 1.1 using ordinary-least squares with both measures of program exposure I then account for the censoring issue by reestimating using a Tobit specification with lower and upper bounds of 0 and 8 respectively, again with both measures The estimated coefficients for de-jure exposure are consistently smaller and insignificant, while the estimated coefficient on de-facte exposure is strongly negatively significant in the Tobit specification While the OLS estimates indicate a drop of 0.3 vaccines on average per year of exposure, the estimated Tobit marginal effect is slightly more negative, at 0.4 fewer vaccines per year For the remainder of the analysis, I focus attention on de-facto exposure

Trang 24

after marriage and stay for extended periods Women who report not being permanent household members account for 473 observations in the data To be conservative, I remove all children whose mothers who are temporary visitors, but this does not meaningfully change the results (Column 5) I also find a small number of permanent migrants, or women who report being permanent household members but who have lived in their present residence for less than the lifetime of the child.in theory, such migration is potentially correlated with immunization outcomes (as it would be if households moved in response to the reform), and may also induce bias In this case, it is not clear that the bias can be a priori signed The data do not permit an examination of the previous residence, but such migrants are limited to only 86 cases Dropping the small number of permanent migrants again

reassuringly does not alter the result (Column 6) The estimated Tobit marginal effect is still -0.4?'

Going forward, I use the smaller sample without visitors and recent migrants Given that the population-weighted mean prior to the reform is 3.37 vaccinations, these results suggest a fall on the order of 12% of the mean per year of exposure

Alternative Specifications

In Table 1.8, J examine other outcomes related to immunization status that are common in the literature Column 1 shows the effect of the program on an alternative measure of immunization status, percentage of age-appropriate vaccines, or the percentage of scheduled vaccinations that are actually received.Further details of the variable construction are in the appendix The Tobit marginal effect based on the table is -4%22

In Columns 2 and 3, I check two other commonly-used indicators, a binary indicator for having received any vaccinations and a second indicator for having received all vaccinations For the latter, I restrict the sample only to children aged 12 months or older, when the official schedule of vaccinations suggests it is appropriate to be fully immunized The probit marginal effects suggest that the program had a negative impact on the probability of receiving both any and all vaccinations, although the latter is imprecisely estimated

Next I verify that the model specification controlling for time/cohort effects using a single wave dummy is valid in spite of its parsimony In Column 4, I include individual birthyear cohort dummies in the specification instead While I strongly reject the joint null that the coefficients on all the birthyear dummies are identically zero (y? = 21.0,p = 0.00), I cannot reject the null that the

coefficients on birthyear dummies for 1990-1992 (NFHS-1) are equal (x? = 2.71,p = 0.26) or the

null that the birthyear dummies for 1996-1999 (NFHS-2) are equal (vy? = 2.11,p = 0.35) The estimated coefficient is also similar to that of the main findings

21 The lack of an effect from potential migration is consistent with limited migration between districts as documented

by Munshi and Rosenzweig (2006), who find that overall migration in India during this period is extremely low and generally confined within a district

Trang 25

In Column 5, for comparison I estimate Equation 1.1 using a different approach, ordered probit regression I treat the variable total vaccinations as an ordered categorical variable, such that more vaccinations are interpreted as “better” The direction of the result is the same - negative and significant In this case, the coefficient on exposure may be interpreted as the fall in the probability of getting one more vaccination as a result of a year of exposure I do not use this method more generally because the coefficient interpretation is not as natural, but it is reassuring to note that the general finding is preserved

In Column 6, I use a Poisson regression where the total time a child is observed is specified equal to the age of the child The coefficient on district exposure in this case is still negative and highly significant This implies that a year of exposure reduces the log of expected vaccines by 0.08, holding all other variables constant

Endogeneous Program Timing

I also consider the possibility that adoption of the program is correlated with observed or unobserved district-level characteristics that, may influence immunization

As a first step, for each district, I compute the delay between the start year and the actual program mandate of 1996 There is no significant correlation between the delay in adoption and key pre-program characteristics like poverty, urbanization, infrastructure and health-related indicators including the infant mortality rate and sample average vaccination rate per district from the NFHS- 1 data In Table 1.9 I show the results of an ordinary linear regression with the years of delay as the dependent variable None of the characteristics are statistically or economically meaningful predictors

In the main analysis, I control for district-level time-invariant characteristics as well as time trends However, I do not control for time-varying differences In the most conservative case, I reestimate the basic specification to allow for district-specific time-trends as well as district-fixed effects by including additional interaction terms with the wave dummy for every district D;*Dg The result, shown in Column 7 of Table 1.8 is weakened but still negative and large, and is marginally

significant at a 10% level (p=0.11)

Survivorship bias

Trang 26

Diversion of Resources to Pulse Polio

Some public health authorities in India have blamed the overemphasis on polio as a result of the national Pulse Polio campaign for the fall in routine immunization.** While I cannot rule out the possibility that Pulse-Polio campaign diverted some resources away from routine immunzation, it is unlikely that the effect of Pulse-Polio is conflated with these particular results: All districts were exposed to the national Pulse-Polio drives equally, while the estimate effect of the program relies on a particular identifying pattern of time and spatial variation

Counterfactuals

In Columns 1, 2, 3 of Table 1.8, I investigate the diflerential impact of the reforms on polio vaccines received, nonpolio vaccines received and the incidence of Vitamin A supplementation If the decline in overall immunization is primarily related to routine immunization as a result of hospital reforms, it would be intuitive to expect less of an effect on theuptake of interventions such as these that are not administered by the district health system While such a result does not constitute proof that the reforms caused a fall in routine immunization, it provides a useful consistency check

Firstly, as previously mentioned, while Polio vaccination continued to be given under routine immunization, the Pulse Polio Immunization (PPI) campaign was implemented during the early 1990s This led to intensive “vertical” campaigns for polio-specific vaccination outreach and camps on National Immunization Days, although some polio vaccines were still administered under routine immunization We therefore expect less of an overall effect for Polio vaccines Columns 4 and 5 shows that the effect is strongest and most significant for the non-Polio vaccines, while the impact on Polio is both smaller and more imprecisely estimated

Vitamin A supplementation to prevent night blindness has also been a national goal since 1976 Biannual doses of vitamin A are given to pre-school children in vulnerable areas While the guidelines suggest that first dose of Vitamin A should be linked to measles vaccination, in practice the program is administered centrally Supplements sponsored by the Ministry of Health and Family Welfare are delivered to rural health facilities directly for distribution by rural health workers During this period, some states also experimented with distributing Vitamin A during the Pulse Polio campaigns (Kapil, 2007) I estimate Equation 1.1 with Vitamin A as a binary dependent variable, and also find no economically or statistically significant effect on the incidence of Vitamin A supplements

Trang 27

1.6 Why Did The Program Hurt Immunization?

The estimated change in household takeup reflects the joint outcome of shifts in supply and demand, but I argue it is more likely to reflect the former The program did not alter the full subsidy for vaccination, and I also control for most of the household and child characteristics that affect demand, such as maternal education, gender and birthorder

The most plausible explanation is that households experienced a decrease in the quality of vaccination services, or rather, an increase in the indirect costs of vaccination In the appendix, I present a simple theoretical framework formalizing the argument that changes in takeup reflect quality changes in the same direction for free services, while only weak increases in takeup reflect quality changes when fees are also imposed

For policymakers in Madhya Pradesh, the decline in immunization is a puzzling unintended consequence of the hospital reforms Even in the presence of the hospital committees, the government specifically chose to preserve overall budget allocations to hospitals for drugs and staff as a floor In particular the centralized procurement of vaccination supplies for each hospital was unaffected

(hence preserving the cost structure)** although hospitals had the option to purchase additional

supplies at their discretion As immunization was exempted for all persons from all fees, the worst case scenario expected was the status quo

However, population coverage is determined both by capital inputs such as vaccination supplies and the labor effort of organizing camps and outreach The previous literature on India documents that immunization coverage is primarily determined by access to public-sector facilities and extremely sensitive to actively conducted outreach such as on-site village camps (Datar, Mukherji, and Sood, 2005; Oster, 2006; Nichter, 1995) There is no record available of doctors’ labor hours, but anecdotal evidence from interviews suggests a reduction in outreach and a reallocation of scarce effort towards the provision of clinical services While the data do not permit a direct test, I present several pieces of indirect evidence that support this story

First, I show that at the hospital level, there was a shift away from outpatient consultation and an increase in delivery and surgical services, in spite of the introduction of substantial fees I also show ‘a similar pattern of decline in the probability of receiving outpatient consultations (prenatal care visits and consultations for children with fever and cough) at the individual level that is correlated with program exposure, with no change in the probability of delivery at a government hospital While it is not possible to separately identify supply-side changes from the decline in outpatient consultations, the (weak) increase in delivery and surgeries in the presence of user fees”? does imply an increase in the quality of these services

Trang 28

An increase in the quality of clinical services need not necessarily imply a reduction in effort on public health, as this may be due to increases in capital equipment purchased with fee revenue and not due to a reallocation of labor efforts However, two other pieces of evidence suggest that effort was in fact reduced in response to the reforms

Firstly, significant and suggestive heterogenity exists between districts I find that the negative

effects of the program are larger in districts with a larger potential fee-paying clientele The negative effects also increase with the realized profitability of the district hospital

Secondly, using a smaller sample of village-level data, I show a reduction in outreach activity that is positively correlated with early program adoption in the rural areas for which data is available

1.6.1 Increased Quality of Clinical Services: Hospital Data

I examine hospital-level data over the 1995-2000 time period collected using retrospective facility surveys in Madhya Pradesh, as centralized administrative records of hospital use are not available While surveys were disseminated to all hospitals,for various reasons only 13 (about one-third) are able to report patient statistics This analysis is therefore subject to many caveats, as the sample is small, likely to be selected on unobservables, and the data cannot be verified independently I choose to report the results as only indicative of changes in doctors’ time use

Table 1.11 shows the change in hospital usage statistics between 1995 and 2000 Over this period, doctors at government hospitals in this sample saw a drop of 20% in outpatient consultations but a disproportionate increase of 30 % in delivery and almost 60% in surgery 2°

I then estimate the impact of the program on the logarithm of inpatients and outpatients and the total number of surgeries and deliveries, using a difference-in-difference approach I regress these outcomes on a dummy for post-program year, a dummy for having adopted the reform early (i.e in 1996) and an interaction term While the sample is too small to yield any meaningful power, the direction of the coefficients in Table 1.12 is consistent with longer exposure to the program increasing clinical services such as surgeries and deliveries, while depressing the number of outpatient consultations?’

1.6.2 Increased Quality of Clinical Services: NFHS Individual Data Next, I use the NFHS data to examine the effect of treatment on the take up of clinical services at the individual level

26Note that at the prevailing average fees, assuming the increase comes from fee-paying services, this generates a net increase in revenue

Trang 29

Maternal Health Services

First, in Table 1.13, I estimate Equation 1.1 on two types of services for which program variation across a cohort allows a measure of individual-level exposure - prenatal care and government hospital delivery Prenatal care, like immunization is largely provided for by the public sector, and after the reform is subject to the outpatient consultation fee However, the NFHS does not identify the location of such care To proxy for hospital location, I restrict the outcome to be a binary indicator

_ for prenatal care by a doctor.I also redefine treatment exposure to reflect the nature of the outcome

variable Instead of a duration measure, I use a binary indicator for whether or not the program is present in the year prior to time of birth, when treatment would be sought In this case, I retain all the previous regressors but omit controls for the age of the child

For delivery, I am able to identify the precise type of health facility, and specify the outcome to be whether or not birth took place in a government hospital For program exposure in this case, I use an indicator for whether the program is present at the time of birth

I find a significant fall in the takeup of prenatal care by a doctor, consistent with the reduction in outpatient treatments(Column 1, Table 1.13) However, there is no significant average effect on the probability of government delivery (Column 3) There are also no significant effects on either outcome in the subsample of the poor only.(Column 2 and 4)

Child Health Services

I also examine the effect of the program on another important outcome: treatment of fever/cough at government hospitals For this exercise, I use only children who have had symptoms in the last two weeks, which is a considerably smaller sample The relevant measure of “program exposure” is unclear I consider the exposure at the district-level, and again adopt a difference-in-differences approach, regressing a dummy for the post-program cohort, a dummy for living in a district that

adopted the reform early and the interaction term?°

In Column 1 of Table 1.14, the results suggest a slight but insignificant increase in the use of government hospitals by children in districts that adopted early In Column 2, I estimate a triple- difference using poverty as a further interaction term find that the results suggest that the effect of early adoption in a districts significantly increased the use of outpatient services the rich by 11%, and decreased the use by poorer individuals by about the same percentage This is consistent with both an increase in quality of services and depressed demand from the poor due to the user fee

1.6.3 Negative Effects Related to Profitability

Trang 30

samples First, I consider the effects in districts that lie above and below the median 1993/4 NSS poverty head count ratio**(Columns 1 and 2).I find that these effects are only large and significant in districts with high overall poverty, where we may expect a larger reliance on the public healthcare system

Each hospital supports immunization activities for the district, while the main revenue base for the hospital committees is clinical services targeted at the urban non-poor If there is a tradeoff between immunization and revenue generation, we are likely to see less immunization when the potential to earn revenue is larger Alternatively, we may think of the program having no possible effects if there is no fee-paying nonpoor urban population To see if the effect on immunization is related to the relative size of the revenue base, I then consider the percentage urban population

(1991 Census) and the urban poverty rate from the 1993/4 NSS Columns 3 and 4 show that the

effect is greater in more urbanized districts Columns 4 and 5 show that it is greater in districts with relatively low urban poverty

Finally I compute a measure of realized profitability from the data available at the hospital- level, by dividing 1999 revenue by the number of hospital beds In Columns 6 to 8, I find that the negative effect is monotonically increasing across the terciles, implying that immunization declines faster when hospital committees are more profitable

1.6.4 Reduction in Village-Level Vaccination Outreach

While detailed information is not available for urban areas or in NFHS-1, NFHS-2 reports the number of family health and welfare camps held in the last year at the village-level for rural areas Such camps are an important source of routine immunization For the NFHS-2 rural subsample, I check to see if the number of camps held in a village is correlated with the length of program exposure in the district The sample contains 123 villages, of which 118 have nonmissing data I group the villages into the early and late adopters, where the latter belong to districts that adopted the reform in 1997 or 1998 Activity in the last year in these villages is thus less likely to affected by the reform

Table 1.8 shows village-level regressions in which the dependent variable reflects the level of

outreach The median number of camps in a given village is, zero The majority of villages (92)

report 0 camps, with a few outliers reporting a maximum of 20 When estimating the effect of early adoption on outreach, I control for the availability of healthcare facilities in each village and other factors that may influence the decision to place camps, such as total population and distance from a town as well as district-level characteristics such as total poverty and health status (proxied for by infant mortality rates in 1991)

Trang 31

the probit marginal effect of being an early adopter on the probability of receiving any camp In

Columns 2 and 3, I report the coefficients from Tobit models with number of camps as a censored

outcome with limits of 0 and 0 and 5 respectively

The results show that villages in districts adopting the reforms more than one year ago are significantly less likely to have had any camps in the last year, and receive fewer camps The average number of camps in the data is 0.2 In both specifications, the estimated marginal effect of early adoption is to reduce the number of camps by 0.2, or effectively removing all outreach*?

1.7 Discussion

This paper provides evidence that decentralization reforms in the health care sector of Madhya Pradesh had an unintended impact on routine immunization - in this particular case, children aged 0-3 years received an average of 12-14% fewer vaccinations per year of exposure to these reforms The paper also suggests that the primary explanation for this fall may be the reallocation of effort away from vaccination outreach towards the provision of clinical services, in spite of policymakers’ intentions to protect against negative effects by preserving pre-reform budget allocations and fee- exemptions for immunization

There are several ways the balance of incentives for doctors may have altered: both before and after the reform, doctors had weak incentives to provide public health outreach, but the reform implicitly increased the returns to providing clinical services While doctors were technically required to perform immunizations and organize camps, they received no performance-related pay Doctors,

moreover, are civil servants with fixed salaries and effective lifetime tenure*! and their hiring and

compensation is not under the committee’s control

After the reform, firstly, doctors received direct non-pecuniary benefits from increasing the revenue of the RKS, in the form of better work environment (Kumar, 2004), and opportunities for advanced training Secondly, although the RKS itself did not have explicit authority to hire or fire doctors, it had a mandate to monitor the doctors and could report back to state officials The ability of the committee to monitor activities, however is considerably stronger within the hospital Thirdly, many doctors in public hospitals also practice in the private sector by operating pathology laboratories or running private clinic services, and are likely to have incentives linked to these outside activities If capital and labor are complements, an improvement in the facilities at the government

3°Note that for consistency with the individual-level regressions, I do not apply state village-level weights in these regressions In Columns 4-6 I present the same analysis, estimated with state village-level weightings While the estimates are no longer statistically significant, note that the magnitude and direction remain similar

Trang 32

hospital may have increased the marginal return to effort in clinic services Younger doctors, for example, may seek to improve their reputations as surgical practitioners for future private practice Established doctors may contract with the hospital to provide services such as laboratory work privately

The lessons that can be gained are naturally limited by the nature of the reform and the available data However, this type of partial decentralization is increasingly found both in and outside the health sector In China, for example, government-owned hospitals have begun to supplement government budgets with user charges and drug sales, but largely leave physician incentive structures unchanged In other sectors, such as education, it is also common to find that teachers remain on a fixed salary structure, answerable to the state, while local schools are allowed to charge fees and determine their own spending

The more general lesson is that incentives matter even when resource constraints appear to be the most pressing concern In the developing country context, Hammer (1997) suggests that the effect of policies on incentives may ultimately outweigh the effect of investment in capital or labor inputs*? Dixit (2002) stresses that, particularly in a multitask setting such as public healthcare, the most important aspect of reform may be the “unforeseen and dysfunctional side-effects of policies” due to changing incentives Indeed, it should be noted that the only documented example of of decentralization that improved vaccination rates directly linked increases in coverage to final compensation (Kremer, Bloom, King, Bhushan, Clingingsmith, Loevinsohn, Hong, and Schwartz, 2006) These findings from India reinforce the idea that policymakers should take potential changes in incentives seriously, in order to avoid creating new problems simply by trying to solve the old

Trang 34

CHAPTER 1 DECENTRALIZATION AND IMMUNIZATION

Table 1.1: State-wise Immunization Rates, 1992 - 1998/9

% Any Vaccine % Fully immunized

Trang 35

Table 1.2: Main Source of Immunization in Madhya Pradesh, 1998/9

Vaccination Source | Urban Rural Total

Public District Health System 82.6% 86.3% 85.4%

Provider Type: Government Hospital 64.3% 20.2% 30.5%

Rural Health Facility 12.1% 23.4% 20.8%

Family health and welfare camp 15.8% 47.2% 39.9%

Other Facility (incl Health Post, Dispensary, Mobile Units) 7.8% 9.2% 8.8%

Trang 36

Table 1.3: Organization of the Rogi Kalyan Samiti Objectives

Improve the hospital, upgrade and modernize the health services Ensure discipline in the institution and supervise the staff Establish affiliations with private institutions to upgrade services Undertake construction and expansion in the hospital building Ensure optimal use of hospital land as per government guidelines Improve participation of the committee in the running of the hospital Ensure scientific disposal of hospital waste

Ensure proper training for doctors and staff

Ensure subsidized food, medicines and drinking water to patients/attendants Ensure proper implementation of National Health Programmes

Ensure proper use, timely maintenance and repair of hospital equipment and machinery

Composition

Management Committee:

Health Minister (President)

Civil Surgeon and Hospital President (Member Secretary) Members of Legislative Assembly (MLA)

District Panchayat President

Related member of Lok/Rajya Sabha Mayor Collector Police Commissioner Chief Medical Officer NGO representatives Press representatives Secretary, Red Cross Society Executive Committee: Collector(President)

Civil Surgeon and Hospital President (Member Secretary)

Trang 37

Table 1.4: Typical Fee Schedule (Rs) 1996-2000: District Hospital, Bhopal

Outpatient 5 per visit

General Ward/Private Ward/ICU 5/68/20 per day Major/Minor Operation ECG Ultrasound Stress Test Endoscopy

Blood & Pathology Tests

Reproductive and Child-Health Related Services

Trang 38

CHAPTER 1 DECENTRALIZATION AND IMMUNIZATION Individual 1992 Survey 1998/1999 Survey N 1868 1936 Age of child (months) 17.2 17.2 Maternal Education HHiterate 0.75 0.71 Literate, some primary 0.04 0.04 Primary Complete 0.09 0.10

Middle School Complete 0.05 0.06

High School Complete 0.05 0.03 University 0.03 0.06 Paternal Education Illiterate 0.36 0.36 Literate, some primary 0.26 0.09 Primary Complete 0.04 0.17

Middle School Complete 0.24 0.14

High School Complete 0.02 0.08

University 0.08 0.16

Maternal age(years) 25.7 25.3

Total household size 8.9 8.0

Total sibling size 2.8 3.0 Rural 0.78 0.76 BPL proxy 0.33 0.29 Scheduled Caste/Tribe (SC/ST) 0.33 0.40 Total Vaccinations 3.61 3.99 Any Vaccinations 0.61 0.81 All Vaccinations 0.21 0.16 Total Polio 1.45 1.92 Total NonPolio 2.17 2.08 Source: NFHS-1 and NFHS-2 Sample weights applied

Định dạng
Số trang	145
Dung lượng	5,95 MB