a; .
may ‘

4.,
a.
.595 ..x..
l. v.
I: /

2 9}} ‘

.1 :1 511

«1.10.4.1... .
12.! 31 .30....) , I.

II z¥kf§iﬂlﬁ

1

L‘ls. ._
.I ’5.

 

I"

3..., m.

.0.

‘(

a}...

A .- 153‘

ﬁnk ngaﬁwlx.

av: » .

3% a

 

 

THEStS

lllllllllllllllllllllllllllllllllllll'llllllllllllllllll

2
d . 31293 01801 7461

 

LIBRARY
Michigan State
University

 

 

 

This is to certify that the

dissertation entitled

The Demand for Child Health Care in 3 Developing
Country: Unconditional Estimates from the Philippines

presented by

Kelly K. Hallman

has been accepted towards fulﬁllment
of the requirements for

PL“ - degree in __Economics_

 

ajor professor

Date December 14, 1998

 

MS U is an Afﬁrmative Action/Equal Opportunity Institution 0-12771

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES retum on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE | DATE DUE

DATE DUE

 

 

Who

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

THE DEMAND FOR CHILD HEALTH CARE IN A DEVELOPING COUNTRY:
UNCONDITIONAL ESTIMATES FROM THE PHILIPPINES

By

Kelly K. Hallman

A DISSERTATION

Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of Economics

1998

ABSTRACT

THE DEMAND FOR CHILD HEALTH CARE IN A DEVELOPING COUNTRY:
UNCONDITIONAL ESTIMATES FROM THE PHILIPPINES

By

Kelly K. Hallman

This study uses discrete-choice econometric methods to examine the determinants
of the demand for curative health care for children in a poor country. It looks speciﬁcally
at the impacts that health care access, price, and quality have on utilization of outpatient
services from different segments of the health care market for infants in one region of the
Philippines.

This research adds to the health care demand literature in several ways. First, it is
relatively unusual in its focus on services for infants and toddlers; this is an important age
to examine as the ﬁrst three years of life are the most crucial period in terms of physical
and mental development, and illness during this time can have detrimental long-term
health effects. Second, in addition to information collected from a broad, multi-wave
survey of households, the study uses a unique set of data on quality of health facilities
and characteristics of health personnel to examine how speciﬁc service aspects drive
demand for care. Third, the study uses BOTH distance and fees for service as measures
of price; most studies in this genre use only distance. Fourth, traditional health providers,
which are used very frequently in developing countries but which are not always
considered in demand studies, are included in the set of modeled health care alternatives.
Fifth, the empirical approach allows differential price and quality responses by type of

health care; most studies constrain price and quality coefﬁcients to be equal across health

care alternatives and assume that the same set of health care characteristics impact the
probability of visiting different types of providers, which greatly limits the insights of
what inﬂuences improving quality or raising fees at different types of facilities will have
on demand. Sixth, and ﬁnally, the model provides estimates of price and income
responses that are not conditioned on self-reported morbidity status; this approach avoids
the selection bias inherent in using a sample of only those who report themselves as ill.

Results from our baseline ﬂexible MN L model strongly reject the equality of price
and quality coefﬁcients across different health care alternatives; ﬁndings indicate that
when differential price and quality inﬂuences are allowed, there are very different within-
characteristic demand responses by type of service.

Choice of health provider for child curative care is inﬂuenced by health care price
and quality, child age and sex, household socioeconomic status, maternal human capital,
and household composition. Distance to services have large negative effects on demand
for care. Controlling for distance, monetary fees do not in general have signiﬁcant
inﬂuences on use. Results for public user fees, however, are quite sensitive to whether
non-health facility community characteristics are controlled for in the analysis. Fees have
strong negative impacts on demand for public care when other community attributes are
either replaced by district-level dummies or omitted altogether; their impacts, however,
are close to zero and insigniﬁcant when these inﬂuences ARE controlled for. This is an
important ﬁnding since results from these types of studies may be used to set pricing
policies for public health care in developing countries: without the detailed community
controls we would conclude demand is price sensitive, whereas with them we would

come to a very different conclusion.

ACKNOWLEDGMENTS

Many people helped contribute to my ﬁnishing this dissertation. First, I am deeply
indebted to John Strauss, for guiding and advising me through the process with patience
and professionalism. His responsiveness and attention to my work were amazing and
very much appreciated; I am fortunate to have been trained by someone of his quality.
Thanks also to go my other committee members, W. Paul Strassman and John Goddeeris,
for their guidance and comments.

Thanks are owed to my friends and colleagues at Michigan State, Kathleen Beegle,
Rosella Gardecki, Leslie Schenk, Alan Barrett for incredible friendship, support, and
understanding. I also wish to thank Agnes Quisumbing and Lawerence Haddad and
IF PRI for providing ofﬁce space and computer support in the latter stages of this
research. Special thanks go to my friends in Washington, DC. for encouragement and
stress relief: Deon Filmer, Signe-Mary McKernan, Jonah Gelbach, Carrie Cihak, and
George Lovell.

Thanks also go to Deon Filmer and Aliou Diagne for discussions on econometrics
and programming; to Jeffrey Rous for helping understand the Cebu data early on; and to
David Hotchkiss and Agnes Quisumbing for providing supplementary data.

Unexpressible love and appreciation go to my mother, Sharon Dove Hallman, for
providing me with the paramount role model of accomplishment and for stimulating my
interest in the world; to my brother, Jerry Dean Hallman, who has always been there for
me and whose inner strength and sense of humor have helped me through difﬁcult times;

iv

and to my father, Jerry Leon Hallman, for providing me with a truly admirable example
of personal renewal and success.

The person I have to thank most, and who I could never possibly thank enough, is
Kirk Deitsch. He is the most unbelievable individual I have ever met. Amazingly, he is
my best friend, most ardent supporter, husband, and lover all in one. He is my ever
ﬂowing well of passion, peace of mind, enjoyment, and intellectual stimulation. The
appreciation I have for his patience, support, and encouragement in helping me realize

this goal I can never adequately expressed. I hope he knows this.

TABLE OF CONTENTS

LIST OF TABLES
INTRODUCTION

SECTION 1

DESCRIPTION AND SIGNIFICANCE OF THE RESEARCH
DESCRIPTION
CONTRIBUTIONS OF THE RESEARCH
SUMMARY OF MAIN FINDINGS

SECTION 2
DISCUSSION OF THE LITERATURE
HEALTH POLICY AND HEALTH OUTCOMES
DEMAND FOR HEALTH SERVICES: METHODOLOGICAL
ISSUES AND RESULTS
AN EXTENSION BEYOND WHAT IS NORMALLY CONSIDERED

SECTION 3
BASIC MODEL OF HEALTH CARE DEMAND

SECTION 4
DATA, SETTING, AND VARIABLES
THE SURVEY
CONCEPTUAL FRAMEWORK: DEMAND FOR INFANT
CURATIVE CARE
PREPARATION OF DATA FOR EMPIRICAL ANALYSIS
DATA EXPLORATION AND CLEANING
CONSTRUCTION OF HEALTH CARE QUALITY AND PRICE
VARIABLES
DEFINITION OF C URA T IVE HEALTH CARE CHOICES

PRICES AND QUALITY OF THE RESPECTIVE FACILITY TYPES
DESCRIPTION OF PA ITERNS OF HEALTH CARE UTILIZATION

C OMM UNI TY-LE VEL NON-HEAL TH FACILITY VARIABLES
INDIVID UAL- AND HOUSEHOLD-LE VEL VARIABLES

SECTION 5

EMPIRICAL MODEL
INTRODUCTION
SPECIFICATION ISSUES

vi

viii

i-‘WOOOO

14
14

17
22

27

31
31

34
35
36

38
46
47
48
49
51

54
54
58

ALT ERNA TIVE-SPECIFIC INCOME COEFFICIENTS

FLEXIBLE HEALTH CARE PARAMETERS

COMPLETE CROSS-EF FEC T S VERSION
ECONOMETRIC METHODS

SECTION 6
RESULTS FOR THE BASELINE POOLED MODEL
ANY VISIT

INDIVIDUAL AND HOUSEHOLD INFLUENCES

HEALTH FACILITY INFLUENCES

EFFECTS OF EXCLUDING NON-FACILITY COMMUNITY

CONTROLS

HEALTH FACILITY CHOICE

INDIVIDUAL AND HOUSEHOLD INFLUENCES
MAIN RESULTS
ROBUSTNESS CHECKS ON HUMAN CAPITAL AND
ASSET RES UL TS

HEALTH FACILITY INFLUENCES
BASELINE MODEL
EFFECTS OF REMOVING NON-FACILITY COMMUNITY
CONTROLS
CONDITIONAL LOGIT SPECIFICA TION
NESTED MUL TINOMIAL LOGIT SPECIFICA T ION
FULL CROSS-EFFEC T S SPECIFICATION

POLICY SIMULATIONS

SECTION 7
RESULTS FOR THE DEMOGRAPHIC AND SOCIOECONOMIC
SUBGROUPS

MALE VS. FEMALE CHILDREN

LOW VS. HIGH MATERNAL EDUCATION

LOW VS HIGH ASSET HOUSEHOLDS

CHILDREN 1-6 MONTHS VS. 7-24 MONTHS OF AGE

SECTION 8

BIBLIOGRAPHY

APPENDICES

vii

58
58
60
61

68
68
68
71
72

77
77
77
81

82
82
85

88
91
94
96
102
103
109
114
119

127

136

145

Table 4.1

Table 4.2

Table 4.3

Table 4.4

Table 4.5

Table 4.6

Table 6.1

Table 6.2

Table 6.3

Table 6.4

Table 6.5

Table 6.6

Table 6.7.1

Table 6.7.2

Table 6.8

LIST OF TABLES

Quality Attributes by Disaggregated Facility Type
Health Care Characteristics by Final Facility Type
Utilization by Demographic Group and Season
Reported Illness by Demographic Group and Season

Exogenous Variables for Analysis of Health Care Demand—Cebu,
Philippines 1983-1986

Summary Statistics--Cebu, Philippines 1983-1986

Any Child Curative Health Care Visit: Individual- and Household-
Level Determinants

Any Child Curative Health Care Visit: Health Provider Attributes,
Included in Succesive Steps

Any Child Curative Health Care Visit: Provider Attributes Included in
Succesive Steps & Community Controls Excluded

Baseline Speciﬁcation of Facility Choice for Child Curative Visit

Facility Choice for Child Curative Care: Individual and Household-
Level Determinants with Barangay-Year Interaction Control Dummies

Facility Choice for Child Curative Care Visit: Provider Attributes
Included in Succesive Steps w Full Set of Community Controls

Facility Choice for Child Curative Care Visit: Provider Attributes
Included in Succesive Steps & Community Controls Excluded

Facility Choice for Child Curative Care Visit: Provider Attributes
Included in Succesive Steps w Community Controls Replaced by
Municipality Dummies

Effect of Facility Price and Quality on Choice—Facility Effects
Constrained to Equality

viii

Table 6.9

Table 6.10

Table 7.1

Table 7.2

Table 7.3

Table 7.4

Table C1.

Table C2.

Table C3.

Table C4.

Unconditional Marginal Facility Effects--pooled Sample Multinomial
vs Nested Multinomial Logit Models

Mean Simulated Probabilities of Facility Choice by Household Asset
Level

Determinants of Facility Choice for Child Curative Health Care For
Male vs. Female Children

Determinants of Facility Choice for Child Curative Health Care by
Maternal Education Level

Determinants of Facility Choice for Child Curative Health Care by
Value of Household Asset Holdings

Determinants of Facility Choice for Child Curative Health Care by
Child Age in Months

Community Inﬂuences for Any Child Curative Care Visit

Pooled Sample — Complete Results for Baseline Flexible
Speciﬁcation of Determinants of Facility Choice for Child Curative
Care Visit

Nested Multinomial Logit — Facility Choice for Child Curative Care

Pooled Sample — Facility Price and Quality Results for Full Cross-
Effects Speciﬁcation

ix

INTRODUCTION

This study examines the determinants of the demand for curative health care for
children in a poor country. It looks speciﬁcally at the effects that health care access,
price, and quality have on utilization of outpatient services for infants and toddlers in a
region of the Philippines. The research is relevant for two primary reasons. First,
because many developing countries have had to cut social spending due to structural
adjustment and stabilization programs which ensued after the debt crises of the 1980's,
new approaches to raising revenues for the provision of health care have taken on greater
importance. Second, young children have not been a primary focus in much of the health
care demand literature.‘ Given that most of the underlying causes of their morbidity and
mortality are from infectious diseases, which are in principle medically treatable or
preventable, improving our knowledge about the underlying factors that most inﬂuence
utilization of basic primary health services for children deserves greater attention.

Many poor countries have large public health care delivery systems; most were
originally designed with the intention of providing universal access; the delivery
approach normally involves two complementary policies: (1) prices for care are set
extremely low or close to zero, and (2) there is no low-income eligibility criteria that

limits access to services by those of higher socioeconomic status (De Ferranti, 1985).

 

' Among those that do examine demand for children's health care are Alderman and
Gertler, 1997; Ii, 1996; Deolalikar, 1993; Gertler and van der Gaag, 1993; Cebu Study
Team, 1992; Dor and van der Gaag, 1987; Wong, Popkin, Guilkey, and Akin, 1987; and
Akin, Grifﬁn, Guilkey, and Popkin, 1986.

Motivations for the direct provision of medical services by government have
several underlying dimensions. Health status is positively related with economic
productivity, and it is one direct indicator of the well-being of a country's population.
Therefore, improving health status is one major rationale for government provision of
subsidized health care. Another motivation is to reduce inequity in access to health care
services. This could be desirable on the grounds that (a) more utilization improves health
outcomes, (b) equity is socially desirable, and (c) increasing access is politically
expedient. Yet another rationale is to help remedy well-recognized market failures
existing in the health sector, namely public goods and insurance market failure, through
the provision of services with public goods aspects and targeting services to those least
able to insure against health shocks--the poor.

Government subsidization with the intention of increasing use of basic health
services is a reasonable goal, particularly considering that overall rates of utilization are
generally low in most developing countries. Whether universal access is the appropriate
vehicle by which to achieve this goal, however, is increasingly being questioned for
several reasons, including the geographical distribution of services that has often resulted
from this approach, the types of the services that normally end up being subsidized, and
because of the severe budgetary reductions for health in most poor countries.

The ﬁrst point is illustrated by data on the distribution of public hospitals and
clinics; the majority of public hospitals and clinics are in urban areas. While this may be
efﬁcient from the standpoint of being able to serve large numbers of people,
concentrating facilities in urban areas may not be consistent with increasing equity in

access or reducing market failures in health care.

 

Concentration in urban areas most likely exacerbates instead of improving
inequities in access because (a) urban dwellers usually have higher incomes--if wealthier
individuals heavily utilize free or low cost public services, subsidies accrue to segments
of the population with higher ability to pay and less ﬁnancial need; and (b) the overall
access costs for care in urban areas are already lower because travel time and travel costs
for are much less because services are closer and more plentiful. Even if public user fees
are uniformly low or zero across areas, total costs faced by urban dwellers are less.

Market failure in the health sector may not be improved because the wrong kinds of
services are subsidized. Social ﬁnancing of preventive services which are not patient-
speciﬁc, such as vector control or sanitation, is justiﬁed for reasons of nonexcludability
of services and external beneﬁt: more than one person is able to receive a unit of the
service at any given time, and beneﬁts of the service, such as reduction in transmission of
pathogens, beneﬁts more than just the direct recipients. Patient-speciﬁc preventive care,
such as child immunizations, while excludable, are candidates for government
subsidization because they create the positive extemality of reduced transmission of
disease. The same holds for patient-speciﬁc curative care for the treatment of infectious
diseases; Because the external beneﬁts of such treatments are not factored into any one
individual's calculation of whether to seek care, demand for such services may be less
than what is desirable from society's point of view. Hence, subsidization is a way to
bolster consumption of externally beneﬁcial types of medical care. The case for
ﬁnancing other kinds of curative care, particularly services not related to preventing the
spread of infectious disease, is weakest because extemalities are much lower. (The

argument usually put forth for government support of these services is that access to

medical care should be a basic right and that ﬁnancial barriers are not socially desirable.
Furthermore, public provision of care is often justiﬁed on equity grounds because it is
viewed as a means to redistribute income (Gertler and Sturrn, 1997).)

Considering that a large amount of public spending on health care goes to patient-
speciﬁc non-basic curative services from hospitals in urban areas, these two sources of
misallocation in public funding-inequity in access and subsidizing inappropriate types of
care--are interconnected and mutually reinforcing.

These problems have led to the suggestion of raising prices for public curative
health care visits, especially in urban areas for adults, whose ailments are less often
caused by infectious disease than are children's. While raising fees is certainly not
expected to remedy all these problems, it could have several positive effects: (1)
allocative efﬁciency could be improved by moving prices closer to marginal costs--price
signals indicate to consumers that they are consuming valuable resources; (2) government
subsidization of adult non-basic curative care for the nonpoor could be reduced; and (3)
net ﬁnancial outlays by the government for health services could possibly be reduced
because fewer expensive non-basic services would be subsidized.

Obviously there are negative aspects of introducing fees for public care too, chief
among them being that access by the poor to modern health care would be reduced.
While other modern providers are sometimes available, they usually charge very high
fees for services. The effects of reducing subsidies for public care depend ultimately on
the sensitivity of utilization to price changes. If individuals are sensitive to small
increases in fees, raising them will generate little revenue and the public health care

system may lose the virtue of helping insure poor individuals against the costs of illness

(Gertler et al., 1995). Furthermore, a uniform increase in fees in urban and rural areas
may worsen already existing geographic inequities in access to care, and demand among
the rural poor could therefore decrease disproportionately.2

The impact of raising fees also depends on other factors including not just own- but
cross-price elasticities of demand, whether public fees are raised from zero rather than
from some already existing positive price, and what the private sector supply and price
responses are to public price increases.

With a government price hike, individuals may opt out of the health care market
altogether; alternatively, they may switch to less preferred types of care. Most health care
demand studies examine the expected results public fee changes will have on demand for
modern public and private care only. Other types of health services are usually not
explicitly considered; traditional health practitioners are used frequently, however, as an
alternative to western medicine in many poor countries. Higher public prices could result
in such services being used more frequently; thus, it is important to consider traditional
health providers as an alternative when examining the effects of policy changes. In
modeling demand for services, traditional care is normally subsumed under the category
of "self-care" instead of being treated a distinct and separate alternative for which prices

are levied for service and whose use inﬂuences health outcomes.3

 

2 Moreover, if much of the curative care that rural poor households seek for their
children is treatment of curable infectious diseases, negative extemalities may result if
utilization for such illnesses falls substantially.

3 Studies that have included traditional practitioners in the choice set are Alderman
and Gertler, 1997; Deolalikar, 1993; Hotchkiss, 1993; Wong, Popkin, Guilkey, and Akin,
1987; Akin, Grifﬁn, Guilkey, and Popkin, 1986; and Mwabu, 1986.

Quality of health services is another critical element affecting demand. Low rates
of utilization observed at public clinics in many developing countries are sometimes
attributed to the poor quality of care offered; poorly trained or insufﬁcient levels of staff
and inadequate drug supplies may inhibit visits to public facilities even if services are
affordable and geographically accessible. This has important implications for the
question of raising fees: if prices are raised when quality is already low, utilization may
drop off even more. However, if fee increases could somehow be accompanied by
quality improvements, decreases in demand may be dampened; raising quality may
enhance both demand for and efﬁcacy of services in terms of improving the health
outcomes of users.4

Knowledge of service quality is crucial for analyzing the determinants of demand
for health services. Because prices and quality are expected to inﬂuence demand in
opposite directions, if data on quality are not available to control for it in our analysis, the
effects on demand of other factors may be biased because they could be picking up

omitted quality. For instance, if price is positively related to the quality of service

 

4 How improvements in services could be ﬁnanced and carried out is another topic
of debate. If quality is to be raised concurrently with or preceding a fee increase,
resources would need to be available in advance to fund these improvements; one
proposal for accomplishing is community-level prepayment schemes. Others have
observed that when fees are increased, low collection rates, poor use of funds, and rising
costs of medical care equipment and supplies often preclude major structural
improvements. Low-cost quality improvements which are probably very important to
patients, such as waiting time, cleanliness, and staff efﬁciency and attitude, may be the
only realistic kinds of upgrades in some areas (Collins, et al., 1996).

Even if funds were available and could be efﬁciently used to enhance quality, it is
not at all clear how revenues would ultimately be affected. This is in part because the
costs of such improvements are not well known; more importantly, however, is the fact
that we still do not have a good idea of how various aspects of quality affect demand for
health services.

 

offered, unobserved positive aspects of a more expensive consultation could result in
price effects appearing to have less of a negative effect than they actually do, or possibly
even being positive. This is a concern not only from the standpoint of academic analysis,
but also for policy making. If empirical results are incorrect due to omitted variables
bias, this could lead policymakers to the conclusion that demand for care is insensitive to
price changes when it may in fact be very sensitive. Those responsible for making
pricing and distribution decision would have the mistaken impression that raising fees
would not signiﬁcantly reduce demand and could increase revenues. Implementing such
policies based on incomplete or incorrect empirical ﬁndings could have unintended
negative results. The take-home message here is that assessing the behavioral changes
expected from a price reform requires knowledge of how both price and quality affect

demand.

SECTION 1

DESCRIPTION AND SIGNIFICANCE OF THE RESEARCH

DESCRIPTION

This research uses a unique set of data from the Island of Cebu, Philippines to
examine the impacts of price and quality on the demand for curative health investments
for young children. The data is exceptional because it consists not only of large,
comprehensive, multiwave household survey, but there is also detailed information on
characteristics of each community, AND on health facilities and health personnel in the
study communities. The household data consists of bi-monthly interviews conducted at
fourteen intervals, covering a period of just over two years. Moreover, the community
and health services data are available both at the baseline period before the launch of the
household survey, and immediately after its completion. Discrete choice models are used
to investigate factors affecting demand for services from modern public, modern private,
as well as from traditional health practitioners. The great breadth and detail of the survey
data allows us to tease out not only how individual and household characteristics
inﬂuence utilization, but also to see what impacts travel time, user fees, and quality of

providers have.

CONTRIBUTIONS OF THE RESEARCH
This research adds to the literature in several ways. First, it is relatively unusual in
its focus on services for infants and toddlers. While we are beginning to understand the

factors inﬂuencing demand for adult health care in poor countries, less evidence exists on

31.1

the

35x

pic

li'ldc

the determinants of utilization for young children; this is an important line of inquiry as
the ﬁrst three years of life is the most crucial period in terms of physical and mental
growth and development (Martorell, 1995). Illness during this period, much of which is
due to treatable infectious disease, can have devastating effects on development because
feeding, appetite, and absorption of nutrients can be severely reduced (Adair, et al.,
1993)

Second is the unique information available on quality of health facilities and the
characteristics of health personnel at the time of the survey. This provides the
opportunity to examine how speciﬁc aspects of services drive demand. The health
service data can be matched with sample households in the communities to augment the
already rich individual- and household-level data.

Third, the study uses BOTH distance and fee for service as measures of price. The
majority of studies use only distance to service to measure price. (This probably based on
the presumption that there are no charges levied for "free" public services, which may or
may not be true. Even if it is, fees are usually always charged for non-publicly provided
care.). If travel-related costs are large relative to the fee for service, it is entirely possible
that utilization of public services may not be reduced as much as might be expected with
a small fee increase at the point of care. This question can only be answered when travel
prices AND user fees are included in the analysis.

Fourth, traditional health providers are a distinct and separate element of the choice
set. It is important from the perspective of designing a health care delivery system to

understand who, when, and for what purposes these kinds of services are used. It may be

wrong to assume that even good quality public care provided at a low price will be used
in all situations given historical and cultural inﬂuences surrounding health and medicine.

Fifth, and ﬁnally, the model provides estimates of price and income responses that
are not conditioned on self-reported morbidity status. This approach avoids the selection
bias inherent in using only those who report themselves as ill. Since there are likely to be
unobservables common to the reporting of illness and the seeking of medical treatment,
using only the group who self-deﬁne as ill is likely to result in biased estimates of the
determinants of health care demand.

Furthermore, the results from the unconditional approach can be considered long-
run in nature. This is because the role that health care utilization plays in the likelihood
of falling ill is factored into the analysis (Dow, 1995a). While we expect prices and
incomes to affect use of care once an illness occurs, it is also true that prices and income
help determine health status itself via their impacts on input choices. For example,
mothers can inﬂuence the probability of child illness by adjusting the mix of health inputs
they provide their children, such as breastfeeding behavior, the timing and composition of
weaning foods, and the use of preventive and curative health care, among others.
Obviously the input mix chosen is inﬂuenced by the prices and attributes of each factor.
Fees and quality of health care, therefore, inﬂuence BOTH who becomes sick, AND
treatment-seeking behaviors once an illness is experienced.

This analysis, therefore, captures the sum total of these health status and utilization
inﬂuences on the ultimate demand for care, while avoiding the selection bias inherent in

using only those who self-report an illness.

10

Results such as these are important for informing policy because they capture not
only factors affecting utilization among the self-deﬁned ill, but they also include in the
calculus who is likely to become ill. Knowledge of unconditional price and quality
responses is in many ways more useful for setting health care policy than responses of

only those who are sick at any particular point in time.

SUMMARY OF MAIN FINDINGS

Our results indicate that long-run demand for child curative services is inﬂuenced
by health care quality, household socioeconomic status, maternal human capital, and
household composition. Distance to service substantially reduces demand for care.
Controlling for distance, monetary fees do not in general have signiﬁcant effects on use.
Results for public user fees, however, are quite sensitive to whether community
characteristics, other than those directly related to health facilities, are controlled for in
the analysis: fees have strong negative impacts on demand for public care when other
community attributes are NOT speciﬁcally accounted for (i.e., when they are replaced by
district-level dummies or when they are omitted altogether). However, they are close to
zero and not signiﬁcant when these detailed inﬂuences ARE controlled for. This is an
important ﬁnding because results from these types of studies may be used to set pricing
policies for care: without the detailed community controls we would conclude demand is
price sensitive, whereas with them we come to a very different conclusion.

Analytically, if community-level health care attributes are included in the
regressions, remaining non-health facility community inﬂuences cannot be controlled for

by using community-level dummy variables because the facility variables would be swept

11

0111; 1
post
lhs
this
carcf

ﬂso

Cun‘

tel

:
v r
IL. I

gene

out; therefore, unless other aspects of the setting are also speciﬁcally accounted for, it is
possible that the health service variables are picking other omitted community inﬂuences.
This implies that results of other health care demand studies, where non-health facility
characteristics may or may not have been adequately controlled for, should be interpreted
carefully; it is possible that even with district-level dummies included, other results could
also be greatly inﬂuenced by non-health-care community inﬂuences.

Strong, though varying, quality effects were found for each variety of curative care.
Current stocks of ORT, the range of vaccines and family planning services available, as
well as the composition of staff were found to have important positive effects on demand
for public services. More sophisticated diarrhea treatments (intravenous) appear to
increase the demand for private care. Demand for traditional care was increased by the
practitioner having had some type of formal health training.

With higher socioeconomic status, households are more likely to use modern
private sector services instead of public and traditional care. Education of the mother
raises demand for modern services generally, and for private, higher-priced, better-quality
care in particular. Paternal education has a similar but smaller effect on private care, but
greatly reduces the demand from services from traditional practitioners. Maternal height
also raises the chances of a private health care visit; this variable is a measure not only of
genetic inﬂuence on child size and health, but it also picks up unmeasured characteristics
related to family background and maternal human capital because it captures investments

by the mother's parents in her health early in her life.5

 

5 In poor countries, nutrition and health investments made by parents during
pregnancy and early in a child's life, particularly the ﬁrst three years, have very large
inﬂuences on subsequent height (Ruel, et a1. 1995) and health (Chen, et a1. 1980).

12

Evidence of differential health investments between older and younger children,
and between boys and girls, is also found. Demand for curative services rises between
birth and six months of age and declines sharply thereafter, even though illness rates do
not decline after six months of age. Utilization rates are higher for male children despite
the fact that their overall morbidity levels do not differ statistically from those of girls.6
Boys are also more likely to be taken to more expensive, higher quality modern private
facilities. Furthermore, additional male infants in the household who are younger than
the index child reduces the likelihood of taking the index child for a curative care visit;
their presence also decreases use of the two more expensive types of care, modern private
and traditional, for the index child. The presence of younger female infants and older
children in the household does not have signiﬁcant effects for the pooled sample;
however, additional male and female infants in residence reduces the probability of a
private care visit for the subsample of female index children.

Additional females in the household over age fourteen increase the chances that the
index child is taken for a private facility visit, even after controlling for household

income and maternal education level.

 

‘5 When split by age, boys have statistically signiﬁcant higher rates than girls in the
ﬁrst year of life, but the trend reverses during the second year when girls have higher
rates. It should also be noted that overall illness is deﬁned as having diarrhea, febrile
respiratory infections (FRI), or any other condition. When disaggregated into the more
well-deﬁned symptoms of diarrhea and FRI, boys have statistically signiﬁcant higher
rates than girls in both the ﬁrst and second years of life.

13

Section 2

DISCUSSION OF THE LITERATURE

Health Policy and Health Outcomes

To motivate the discussion of the literature, it seems reasonable to ask why it is
important to examine the determinants of health care demand in the ﬁrst place. One
reason is equity in access; this was discussed earlier as one of the primary motivations for
government subsidization of health services. Policy levers can make access to health care
more even by lowering fees, reducing travel time to care, and improving the quality of the
services used by the poor. Knowledge of how these policies inﬂuence demand across
different segments of the population is needed to judge their effectiveness in reducing
differentials in access.

Second is that a ﬁnancing agenda for the health sector can be appropriately
designed only when it is known how prices and quality affect utilization (Gertler and
Rose, 1997). Willingness to pay for different types of care and the various attributes of
those services determines which improvements in the system should and could be
reasonably undertaken; it also helps guide the choice of ﬁnancing schemes.

Third is that health care is expected to inﬂuence health outcomes. While this

 

proposition is usually accepted as common knowledge, the evidence is not plentiful for
several reasons. First is that measuring health status is difﬁcult, as evidenced in the
earlier discussion of self-reported measures. Second, use of health care is likely to affect
health outcomes slowly, making its beneﬁts difﬁcult to capture. Moreover, many health

care policies, such as prices and quality, inﬂuence health status only indirectly through

14

5110

p051

113m
111m
in Ci
tr111t-

COmn

their effects on health input choices. Because input choices are affected by many other
factors aside from the characteristics of health care, such as preferences, prices of other
goods, incomes, etc., the effects on health directly attributable to health policy changes
may not be of ﬁrst-order magnitude.

There exists evidence at the micro-economic level of positive effects of health
service availability and quality on health outcomes. Using the 1988 Ghana LSMS, Lavy,
Strauss, Thomas, and de Vreyer (1996) ﬁnd that availability of child health services
offered in the community increases child survival rates; characteristics of health
personnel are signiﬁcantly associated higher with child height. For rural children,
immunization services and drug stocks improve anthropometric measures, while price
and distance to services have negative effects. Results for the health of adults is not as
strong, probably due to the high mobility of the population; only drug availability had a
positive and signiﬁcant impact on adult BMI and only in urban areas for women.

Using this same Ghanian data, Benefo and Schultz (1994) ﬁnd higher drug prices
increase child mortality. Thomas, Lavy, and Strauss (1996) ﬁnd drug supplies and
immunization services are associated with an increase in child nutritional status (height)
in C6te d’Ivoire, particularly among households with no education. Other studies that
have investigated the effects of health care availability and quality on child health
outcomes and survival‘ include Barrerra (1990), who ﬁnds an inverse relationship
between distance to a clinic and child height in the Philippines; Strauss (1990) ﬁnds

community perceptions of problems with local health services to be associated with lower

 

‘ These are cited in Lavy, Strauss, Thomas, de Vreyer, 1996.

15

child height and weight for height in rural C6te d'Ivoire; Rosenzweig and Schultz (1982)
show that child mortality is lower in urban areas where there are more clinics and
hospitals. Thomas and Strauss (1991), on the other hand, ﬁnd in Brazil that child height
is inversely related to the number of nurses and beds in the community.2

Moving from the micro- to the macro-level, the ability to associate health outcomes
with health sector policy is not as strong. Comia and Mwabu (1997) use country-level
panel data to analyze the determinants of changes from 1960 to 1995 in infant, under-
ﬁve, and maternal mortality rates of forty Sub-Saharan African countries. They ﬁnd,
controlling for time trends, that calorie availability, female literacy rate, per capita
income, safe water supply, immunization coverage, and broad access to health care3 all
have negative impacts. The importance of each regressor, however, varies according to
the speciﬁc group considered; access to health services had noticeable impacts only on
infant mortality rates.

Filmer and Pritchett (1997) use cross-national data on under-ﬁve and infant
mortality rates from 100 countries to examine the impact of nonhealth factors and public
health spending on health. They ﬁnd that the vast proportion of cross-national variation
in mortality can be explained by the country's per capita income, the extent of women's

education, the level of ethnic fragmentation, and the predominant religion; public health

 

2 This could be due to purposive placement of facilities in areas where children are
sicker and less well-nourished. This would make the presence of facilities appear to have
a negative effect on child health, when in fact, initial poor child health was precisely why
the facilities were placed in the area to begin with.

3 The health care variable is the percentage of the population having access to
health services; however, the deﬁnition of "access" is not speciﬁcally given in the text.

16

spending has only a very small impact. Potential explanations for this non-result on
public spending, the latter three of which are acknowledged by the authors, include (a)
the fact that only part of health care consumption is represented by public spending:

large amounts are spent by private individuals on the consumption of private and
traditional services; furthermore, sizeable individual costs (travel costs, travel and waiting
time, and sometimes user fees) are incurred even in the consumption of public services;
(b) public spending is not always allocated to the most cost-effective services in terms of
improvement in health status per dollar spent; ( c) additional spending on public health
care may merely substitute for private services that would have been consumed anyway,
so there is no net increase in total health care consumed; and (d) more spending does
necessarily represent increased consumption of services because of variations in the
efﬁciency of service delivery: a dollar of public spending in one country may translate to
more services, while in another it may not.

Another potential issue with both of these studies is that mortality is the health
outcome used. While this indicator has the advantage of being widely available, and is
less subject to measurement error and reporting bias than morbidity, it cannot capture
potential health improvements, which may stem in part from consumption of health

services, experienced by those who remain alive.

Demand for Health Services: Methodological Issues and Results

17

res;

OTIS

1216.2

The empirical work on health care demand focuses on the impact of price on
demand for services from various types of providers using discrete choice models.4 Some
studies have found that price decreases demandf while others have shown demand to be
price inelastic.6 The differential results could be attributable to a variety of factors: (1)
the price of health care is not deﬁned in the same manner across studies; (2) different
responses by socioeconomic groups are not always allowed for; (3) quality of services is
often not explicitly taken into account, and if it is, limited and sometimes unreliable
measures are used to proxy for it; and (4) the data come from different countries, settings,
and periods of time. The ﬁrst three issues are important from the standpoints of
informing future data collection efforts and the choice of analytical methodology for

health care demand studies, and therefore merit further discussion.

 

4 Other studies monitor the impacts of actual policy changes by observing
utilization before and after a price increase. In most cases overall utilization is lower
after the price increase, and it decreases more in poor areas. See Appleton, et al. 1996;
Collins, et al. , 1995; Creese and Kutzin, 1995; Kutzin, 1994; Creese, 1991; Lewis,
1991a, b; Yoder, 1990; Santon and Clemens, 1989; Waddington and Enyimayew, 1989.
One study, however, by Litvack and Bodart (1993), ﬁnds that utilization rises after a
price increase; this is presumably due to concurrent quality improvements.

5 Alderman and Gertler, 1997; Gertler and van der Gaag, 1990; Mwabu, 1989; and
Gertler, Locay, Sanderson, 1987.

6 Schwartz, Akin, and Popkin, 1988; Akin, Grifﬁn, Guilkey, and Popkin, 1986;
Birdsall and Chuhan, 1986; and Heller, 1982.

18

The price of care faced by households consists of user fees, travel costs, travel
time, and time spent at the facility.7 These price variables are the most immediate policy
levers available to inﬂuence utilization; evaluating the impact of price on demand,
therefore, means that clearly deﬁning price is crucial. Often the only data on prices is
household expenditure per medical care visit; this measure is not exogenous however
because it reﬂects choice of provider which is likely to be inﬂuenced by variation in
service quality (Deolalikar, 1993; Deaton, 1988) as well as individual and household
level characteristics such as education and income. (For example, there may be price
discrimination by socioeconomic status.) If expenditure is the only information available
on price, one approach to minimize the endogeneity of this variable is to construct "unit
prices" for each type of health service; this is done by aggregating per visit expenditure
by provider and service type, e. g., a visit at a public clinic for basic curative care for a
child. The "unit price" for this type visit is then deﬁned as the median or mean value
within a small exogenously-deﬁned geographic area, such as a community or sampling
cluster.8 If available, price information from providers may be better than household
expenditure data because it is less likely to be inﬂuenced by household characteristics; it
may also reﬂect the price of what a provider considers a typical visit for a speciﬁc type of

service offered; e.g., a routine visit for a child with diarrhea or a respiratory infection.

 

7 Time spent traveling, waiting for, and receiving services are opportunity costs in
the sense that the time could have been used in productive activities or in leisure. The
value of these time costs has often been approximated by the hourly wage the individual
would receive had he or she been working for pay instead.

8 Cell sizes of at least 10 observations are normally desired for this procedure in
order to avoid the undue inﬂuence of outliers.

l9

The accuracy of facility data may itself be a concern, however, because ofﬁcial fees may
not be the same as those actually charged (Thomas, Lavy, Strauss, 1996); providers may
offer better or more expeditious care to those willing to pay higher prices. For instance,
fees may be accepted for better "free" services. The ofﬁcial fee data will show no
variation in prices, when in reality there could be.

Obviously, lack of variation in a variable is problematic for analyzing responses to
it. Even with zero or very low user fees, however, other elements of price, mainly travel
time and costs, will vary across households and communities; furthermore, they are often

a very large part of the total price of obtaining care.

nt u
In terms of different responses by socioeconomic level, several studies ﬁnd demand
to be more price elastic for lower-income households.9 These results suggest that fee
increases may affect the poor more; demand also appears be more elastic at higher price
levels, implying there is eventually a diminishing revenue return to increasing prices. It
is important to note, however, that price was not deﬁned the same way across these

studies. If travel time (i.e., distance) is used as the measure of price, and its magnitude is

 

large relative to the monetary fee for service, it is entirely possible that utilization of
public services may not be reduced as much as might be expected with a small fee

increase for service.

 

9 Alderman and Gertler, 1997 ; Bitran, 1993; Ellis and Mwabu, 1993; Gertler and
van der Gaag, 1990; Dor, Gertler and van der Gaag, 1987; and Gertler, Locay, and
Sanderson, 1987.

20

Wis:

The effect of quality on demand has long been recognized as potentially important,
but only recently has it begun to be speciﬁcally addressed. One obvious reason many
past studies did not include it was lack of data. Alderman and Lavy (1996) review a
handful of recent studies that use both household- and facility-level data to examine
household responses to cost and quality of services. The results, mostly from African
countries, indicate that distance to facilities and availability of basic drugs are the two
strongest inﬂuences on the demand for adult curative care. A study of obstetric care
choice from the Philippines (Hotchkiss, 1993) ﬁnds distance, waiting time, and
availability of doctors to be signiﬁcant determinants of choice of facility for child-
delivery services; drug availability had effects only among more educated women.

Several of these papers performed simulations of demand responses to improving
public services. They indicate large potential increases in utilization; much of the likely
increase reﬂects switching from private into public care, although there also appears to be
slight shifts from self care and traditional health care into public modern services. An
important question is whether such improvements could offset the negative effect of fee
bikes on demand. One example from Ghana (Lavy and Germain, 1994) ﬁnds that when
quality enhancements were accompanied by fee increases in the simulations, demand for
quality appeared to be high enough that doubling the value of several of the quality
indicators simultaneously would not decrease overall demand at public facilities unless
prices were increased by more than twelve times their original level.

Better quality ﬁnanced through higher fees may not have equally beneﬁcial effects

across all socioeconomic strata however. The few analyses that have estimated

21

ii dill,

mom:

who r:
ptopl:

mobs

morbj
magi
healt'r

FOYE‘.

mile:
llltvrr
0f illr
“18> l
Objec
€011.51

differential responses by income, have shown willingness to pay for improvements such
as drug availability, facility infrastructure, personnel, and range of services, rises with

income.

An Extension Beyond What is Normally Considered
U ' ° ' ate

Finally, most studies estimate demand for health care only for those individuals
who report a current illness. Conditioning on illness makes some intuitive sense: healthy
people will not demand curative services. However, selection bias is an issue if
unobservables related to the self-reporting of illness also affect demand for care. Factors
associated with seeking care during illness, both observed and unobserved, may also
inﬂuence the self-reporting of the health status.10 If such reporting biases were correlated
with only observables, then that would not lead to biased estimates of short-run effects.
The problem, however, is often one of unobserved attitudes toward care-seeking and
morbidity. If these attitudes do not change as observables, such as income, change, then
marginal income effects from conditioned estimates will be biased because self-reported
health status will be correlated with the error term of the health care demand equation.

For example, those who tend to under report illness may also tend to avoid modern health

 

'° An example from Peru shows the relationship between maternal education and
maternal-reported incidence of child illness follows an inverted-U shape (Sindelar and
Thomas, 1991). If more educated mothers have better information and greater awareness
of illness symptoms, perhaps because of more experience with health care providers, they
may be more likely to report their children as sick (Strauss and Thomas, 1995). More
objective measures of health and nutrition, such as child anthropometric status,
consistently show positive effects of maternal education.

22

care when sick;“ conversely, a person with unobservably poor health may be more
familiar with the health care system and be more likely to demand care when sick (Dow,
1995a; SChultz and Tansel, 1993; Dor and van der Gaag, 1993, 1987; Ellis and Mwabu,
1991)

Another reason health status is endogenous for the empirical estimation of health
input demands is that individuals are to some extent able to choose their expected level of
health through their choices of health inputs (Dow, 1995a). Therefore, over the long
term, prices and quality of curative services inﬂuence not only demand for those services,
but health outcomes as well.

The implication is that estimated responses to health prices which are conditioned
on a self-reported current illness may be biased due to self-selection into reporting illness.
Furthermore, they may only be valid in a short-run sense because they do not factor in
how long-run health care demand is inﬂuenced by the mix of health inputs chosen which
in turn affect health status. Unconditional estimates are free from the selection bias
arising from morbidity self-reports; they may also better capture the effects of policy
changes, such as fee increases, because they include not only the short-run demand
effects, but also the longer—run feedback effects that reforms are likely to have through

their impacts on health status.12

 

“ For example, an individual may currently be healthy and not demand curative
services; however, when an illness does occur, if he/she forgoes treatment, health may be
worsened in subsequent periods, and demand is actually increased in the long-run.

‘2 Dow (1995a) points out that estimated long-run effects may themselves be
difﬁcult to interpret if changes in health prices or quality over time are unobserved.

23

In order to estimate both conditional and unconditional responses, both illness and
health care utilization data must be available from all sampled individuals and for the
same recall period.'3 In a study with such information from Cote d'Ivoire, short-run
elasticities from the conditional, self-reported ill-subsample were found to be
approximately twenty-ﬁve percent larger (i.e., more negative) than those estimated for the
entire, unconditional sample (Dow, 1995a). While demand for most goods is thought to
be more elastic in long-run because of substitution opportunities, this cannot be
determined a priori for health care investments. It is possible that, regardless of reported
health status, current prices reduce quantity demanded, making health worse enough in
subsequent periods that quantity demanded is driven back up. Alternatively, with given a
price increase for both preventive and curative care, preventive may not be as attractive a
substitute for curative care.14 This could have negative impacts on health, thus increasing
long-run curative demand.

This paper also found signs and significance of income, age, and gender variables
differ between the conditional and unconditional estimates. The estimated wage had
signiﬁcant negative impacts on hospital utilization when conditioned on an illness being
reported, but no inﬂuence in the unconditional estimates. Age had signiﬁcant negative

effects on demand for hospital and clinic care in the conditional sample; however its

 

'3 In this data the health care utilization questions were not preceded by screening
question "Was the individual ill in past two months?" We know, however, whether the
visit was for curative or preventive care. Unfortunately, the health data is not available
for the same recall period as the health care data. Therefore, conditional demand cannot
be estimated and compared with the unconditional estimates.

‘4 This assumes that the same facility is used for both preventive and curative care.

24

inﬂuence was positive and signiﬁcant for both types of care when the estimates were not
conditioned. For those reporting an illness, being male had a positive inﬂuence on
hospital, but not clinic visits; however, using the unconditional sample the male effect is
strong and negative for clinic visits, but has no signiﬁcant impact on demand for hospital
care.'5

Using data from Indonesia, Deolalikar (1993) estimates the probability of seeking
care from modern and traditional providers, ﬁrst conditional on an illness having been
reported and then for the entire unconditional sample. Contrary to Dow's results, he ﬁnds
long-run own-price elasticities to be larger (i. e., more negative) than those ﬁom the ill
subsample. One possible explanation for these results is that the sickest of the sick are
more likely to report an illness; if these individuals consider themselves to be in dire need
of medical care, perhaps they are less responsive to price than others who have not
deﬁned themselves as ill. This would lead to downwardly biased estimates (more toward
zero) in the conditional price elasticities. This does not, however, reconcile the
differences in the ﬁndings between the two studies.

One cross-price effect differed between the long- and short-runs: modern service
prices signiﬁcantly increased demand for traditional services in the short-rim, but not in
the long-run.

Changes were also observed on key household and community variables. For

modern services for children, schooling of the household head signiﬁcantly increased

short-run demand, while in the long-run its inﬂuence was negative and non-signiﬁcant;

 

'5 Dow (1995a) demonstrates that long-run, unconditional elasticities are equal to
the sum of the short-run elasticity of curative care, plus the elasticity of health with
respect to the covariate of interest.

25

LUNG

etlec

m;

111 lit

child age had no noticeable short-run effects, but was strong and negative for the
unconditional sample; organized trash collection in the community had positive but small
effects in the short-run, whereas in the long-run the inﬂuence was large and negative.

While these two studies use data from different areas, and do not deﬁne price
variables nor handle certain speciﬁcation issues in the same manner,‘6 the dissimilarities
in the ﬁndings call for further research into the nature of long-run demands for health

services.

 

‘6 One speciﬁcation difference is that Dow allows varying degrees of cross-
alternative substitution by using nested multinomial logit and multinomial probit
structures.

26

Section 3

BASIC MODEL OF HEALTH CARE DEMAND

The model presented here is a household production model for health inputs and
outcomes; it is similar to that used in previous health care demand studies. It is assumed
that the household maximizes a utility function, the arguments of which consist of health
of the infant (H) and consumption of a composite good (G), conditional on (Z) which is a

set of taste and preference shifters”:

(1) U=U(H, G; Z)

Health of the sample infant is produced by combining inputs in the manner implied by the
health production function. This function is modeled as a relation between the health
outcome and a set of health input choices; its shape will depend on the underlying health

technology. The production function is written

(2) H=H(C,F;S,M,E,u)

 

'7 Such a unitary model of household decision making, in which households are
assmned to make decisions which maximize household utility, does not allow one to
delve into the processes of intra-household decision making. The unitary approach is
used because information is not available in these data on individual non-labor income or
individual ownership of assets within the household. Information is available, however,
on characteristics and health-related behaviors of mothers of the sample children.

27

where the ﬁrst two arguments are endogenous inputs into health: C is the quantity and
quality of health care chosen and F consists of other health inputs, such as food and
nutrient intakes and health-related behaviors such as cooking, food storage, santitation
and excreta disposal practices. S, M, and E are exogenous characteristics that inﬂuence
infant health: S is the set of individual attributes of the child such as age and gender; M
consists of household characteristics including age, education, and family background of
the the child's parents, and E is the set of community characteristics which inﬂuence
health, such as sanitation, water quality, rainfall, temperature, and the general disease
environment. It should be noted that S, M, and E can have both direct effects and indirect
effects through C and F. 1) represents child- and household-level unobservables such as
inherent healthiness of the child.

The household also faces a budget constraint:

(4) Y=PcC+pFF+G

where Y is household income, pC is the price of health care, and pF are the prices of other
health-related inputs; the price of the composite good is normalized to one. The price of

health care is comprised of the user fee and access costs such as travel time to the facility.

(5) pc=13+WT

where B is the user fee, w is the wage rate, and T can represent travel time to or waiting

time at the facility. Substituting (5) into (4) gives the full-income budget constraint

28

(6) Y=(B+wT)C+pFF+G

Substituting (6) and (2) into (1) gives the conditional utility function for health care
choice j
(7) UJ- = (Ill-(C1, Ff"; S, M, E, u), Yj - BJ- Cj - wTJ-Cj - pFFj“; Z)
where F f" is the optimal choice of other health inputs given health care choice j.
To specify the utility maximization problem for choice of health care, suppose the

individual (the child's mother) faces J feasible health care alternatives. The unconditional

maximization problem is

where U* is maximum utility. The solution to the utility maximization problem gives the
health care alternative that is chosen. When stochastic terms are added, the probability
that an alternative is chosen can be interpreted as a demand function in a discrete choice
model such as the one speciﬁed here.

It should be noted that the processes and dynamics of health production are not
taken into account in this analysis. It is assumed, however, that (1) inputs chosen in
previous periods affect current health, and (2) health in the previous period affects current
health. These assumptions imply that in a dynamic model both lagged and expected

future values of exogenous variables would enter the reduced-form demands. In the

29

empirical work, several covariates enter with current and past values (e.g., rainfall and
food prices), others are time-invariant (e. g., parental education), and the remainder are
assumed to change slowly over time (e. g., health care availability and quality). The very
young age of the children in the sample, and hence the short time-period over which their

existing stock of health is based, makes these assumptions more tenable.

3O

Section 4
DATA, SETTING, AND VARIABLES

THE SURVEY

Household, community, and health facility data from the Cebu Longitudinal
Health and Nutrition Study are used for this study. The Cebu data are a rich resource for
examining issues related to child and maternal health. The survey period, 1983-1986,
coincides with the introduction of structural adjustment and stabilization programs and
severe economic downturn in the country. Unemployment, inﬂation, and poverty
increased during this period; nutrition, health, and education indicators also worsened
(Herrin, 1990, 1992). The region to which Cebu belongs saw the proportion of
underweight children increase during this time; by 1987 this area had the highest
prevalence of low weight-for-age children in the country. Furthermore, in this particular
sample, half of the children at the age of two years had heights two or more standard
deviations below the WHO reference median for their age, suggesting a high prevalence
of chronic under nutrition (Glewwe, King, Adair, Jacoby, 1994).

The site is Metropolitan Cebu, an area in the central Philippines, which includes
Cebu City, the second largest city in the country, and surrounding urban and rural
communities. The area is located on the eastern coast of Cebu Island and includes a
number of coastal, island, and high elevation villages which vary in environmental,
socioeconomic, and agroecological conditions. Following an initial pilot survey in 1982,
seventeen of the 158 urban, and sixteen of the 85 rural, barangays (villages) in the area
were randomly selected to be included in the sample. The sample consisted of all

pregnant women and their households in these 33 sample barangays who could have

31

delivered a single child between May 1, 1983 and April 30, 1984. Baseline pre-birth
surveys were conducted with the 3327 women who ﬁt this criteria. Subsequent
interviews were conducted immediately following each woman's delivery and then every
two months through the ﬁrst two years of each index child's life. A few women were lost
to the sample immediately after the baseline survey and a handful more following the
post—delivery survey due to out-migration, twin births, stillbirths, miscarriages, and
refusal to be interviewed. At the beginning of the bi-monthly longitudinal surveys, the
sample consisted 2884 woman-infant pairs.18

The household surveys collected information on household composition; human
capital of household members; value and composition of household asset holdings (at the
baseline and ﬁnal longitudinal round); household income and detailed information on
labor supply and sector of employment of the index child's parents (in survey rounds I, 6,
and 12); household sanitation conditions; health insurance coverage; mother's
contraception behavior and fertility history; infant feeding practices; maternal prenatal
behaviors during her pregnancy with the index child (at baseline); type of practitioner
used for child delivery services (post-birth interview); mother's postpartum health care
utilization; and health care utilization for each index child.

Besides the 3327 households, eighty-two modern health facilities, mainly public
and private hospitals and clinics, used by the sample population, were also surveyed. A

broad range of information was collected on types and number of personnel, drug

 

‘3 The mean number of completed longitudinal surveys per individual was 10.5 out
of a possible 12. Missing surveys were due to migration, withdrawal from the sample,
and a few infant deaths.

32

availability, hours of operation, kinds of services offered, prices, and intensity of patient
use of the facility. Two separate health personnel surveys, which covered both modern
and traditional practitioners, gathered data on education, training, knowledge, and
attitudes of individual health providers.

In addition to the household and health facility surveys, information was collected
on a wide range of characteristics of each sampled barangay. This survey, conducted at
the time of the baseline household survey and again after the completion of all the
household surveys, contains information on population, water and sanitation
infrastructure, other physical infrastructure (including roads, availability of
transportation, electricity, and other services), the agroecological setting, prevalence of
local community groups and organizations, and the presence of different types of health
and educational institutions as well as commercial and retail establishments and
activities.

Community-level market food prices were gathered at ten equally-spaced
intervals during the survey period. These covered a large number of food items. For
each type of commodity, prices were collected on several different brands so that a range
of qualities for each product was represented. Furthermore, careful attention was paid to
measuring the units, weights, and volumes of each product. This ensured that price per
standardized unit could be derived for each food item.

Monthly rainfall levels for the area from 1973 to 1986 were also available;

information on current and rainfall lags up to two years were used in the analysis.19

 

‘9 The author would like to thank Agnes Quisumbing for sharing this data.

33

Current and lagged rainfall levels are expected to be important determinants of morbidity

and income and therefore to have large effects on health care utilization.

CONCEPTUAL FRAMEWORK: DEMAND FOR INFANT CURATIVE CARE

The demand for health care and other health inputs is a function of many different
factors. These can be grouped loosely into the categories of preferences for health
relative to other goods, the household's budget constraint, ﬁxed inputs, and the child's
health endowment.

Preferences are associated with observed and unobserved household
characteristics. Observed characteristics could include education, sex, and age of the
health care decision makers (e.g., the child's parents). Factors inﬂuencing preferences
which are normally unobservable could include tastes for higher-quality health care;
perceptions of child health care needs; beliefs concerning appropriate types of health care
for children of differing ages and sex; the seriousness with which child illness is viewed;
inter alia. Many of these could be related to education, cultural or ethnic afﬁliation, and
family background.

The budget constraint is determined by the household's income; the prices it faces
for health care such as user fees, transportation time and costs, waiting times, etc.; prices
for other health inputs such as food; prices for non-health consumption goods; and the
market wage (i.e., the opportunity cost of time).

Fixed inputs which affect health input choice can be at the household or
community level. At the household level these could include education, age, and family

background of the mother. Better educated mothers, for example, may not only have

34

stronger preferences for health utilization, they may also make better decisions regarding
choice of health inputs.

The child's health endowment reﬂects the innate healthiness of the infant. This is
expected to be an important inﬂuence on child health care choices made by the mother;
this factor, however, is not directly observable to the researcher.

The choice of which health inputs to use and the amount of each to use depends
on all the exogenous variables that are observable to the researcher and those which are
unobservable. The relationship between the observable exogenous factors and the
demand for health care is the health care demand function. (The derivation of demand is

laid out in more detail in the modeling section).

PREPARATION OF DATA FOR EMPIRICAL ANALYSIS

We now turn to a discussion of the data, starting with a general description of
activities related data exploration, quality checks, and cleaning. We then discuss the
construction of the health care characteristics data, followed by an explanation of how
the health care choice set used in the analysis was determined. We then describe health
care utilization patterns of the households at the time of the survey. This is followed by a
discussion of the exogenous explanatory variables at the individual-, household-, and
community-levels used in the analysis; we then present summary statistics for the

regressors used in the analysis.

35

DATA EXPLORATION AND CLEANING

In general, construction of the data sets used for the analysis was labor-intensive
and time-consuming because of the complexity, size, and breadth of the survey. The
survey covered 3327 sample mother-infant pairs and their households. Each of these
sample pairs was surveyed 14 times, once at baseline, once just after delivery, and at
twelve bi-monthly longitudinal rounds. This resulted in approximately 42,000 total
person-observations in the sample. For each person-observation there was a vast variety
of information collected on the child, the mother, and their household of residence. In
addition, data was collected on the general attributes of each of the 33 barangays
sampled; furthermore information was also gathered from each health facility used by
sample mothers and infants.

With a data set of this size, the ﬁrst activity undertaken for empirical research is to
explore the ﬁles for basic content. This includes investigating which speciﬁc variables
each ﬁle contains. It is also crucial to know how these variables were actually measured.
For instance, non-labor income in one survey could mean just remittances received, while
in another it could also include interest on savings, remittances given, pension income,
etc. It also important to know how each question was phrased to ascertain whether the
desired information was collected in an accurate, unbiased, clear manner. In a multiple
round survey, it is necessary to explore how frequently each type of information was
obtained and for what recall length, and if the question was asked in the same way each
time. If a question was asked at every survey round, each different temporal observation

is matched to the base individual for each period. On the other hand, if a different piece

36

of information was asked only once in the survey, this single value will be attached to
that individual for every round.

In general, matching records ﬁom surveys with multiple observations not only on
children, but also on households and communities is not straight-forward. It is critical
that, to the extent possible, errors are identiﬁed and the data matched correctly. It is not
uncommon, for example, in these sorts of surveys for an individual's identiﬁer to change
across rounds. At the household level, questions are often asked of each individual in the
household. For analysis this means that for each of these person-records, variables had to
be created and then combined into one single record which is tied to the index child. For
community-level data, each relevant piece of information is attached to individuals
currently residing in that community. Another issue with community-level data is that it
can be collected at different times, at different intervals, and for slightly different
geographic boundaries than the individual and household data is. This implies that
various levels of matching are required to get the correct value attached to the desired
person in the appropriate village and for the right period of time.

For discrete categorical variables in a repeated survey, it is important to check that
the response codes are the same across rounds. Often after the ﬁrst round of data
collection, the researcher often learns more about the particular setting he/she is working
in which, and this calls for changes in coded responses to better reﬂect the setting. If
responses are added or changed, it is important to know whether and how these can be
matched to the previously used codes so that responses for each period of time can be

used in a consistent manner.

37

In terms of data quality, it is important to check outliers to see how they compare
against values for other similar individuals (or households) in the ﬁle for same period of
time, and against values for that same individual or household over different rounds.
Sometimes obvious mistakes in data entry can be detected in this manner.

Rules checks are another important category of data cleaning. One type of rule
are nested "if, then" type of questions. With an afﬁrrnative response to a screening
question (such as whether health care was sought), if questions within that nest have no
values attached to them, their values should be set to missing and not to zero. Another
type of rule is the consistency of mathematical or accounting relationships. For example,
a child's age in any survey round should be the difference between his/her birth date and
the survey date (many mistakes were found in this variable in our data). A ﬁnal type of
rule relates to conditional relationships between different variables. For example, an
individual coded as being the mother of an index child should not be coded as being male
in gender.

It was tedious, but very important, that these types of subtleties and errors be F
examined before doing the analysis. Furthermore, it is crucial that each change be

documented in a program ﬁle for future reference.

 

my
‘1

l .

CONSTRUCTION OF HEALTH CARE QUALITY AND PRICE VARIABLES

The Cebu data contain a large amount of information on health facilities. Two
surveys of all modern facilities used by the sample households, both inside and outside
the sample barangays, were conducted for the study, once at baseline and once near the

end of the survey in 1986. Information on prices for different types of services, the range

38

of health services offered, and stafﬁng levels were collected at both rounds. Data on drug
stocks in facilities were available in both surveys (the drug information contained in the
1986 survey, however, was much more complete). Two health personnel surveys were
also administered; these gathered data on education, training, beliefs, and practices of a
wide range of staff in modern facilities, as well as traditional providers and midwives.
Ofﬁcial service fees were collected from modern providers; however, no user fee data

was collected from traditional practitioners.

EealthQaLLQualjnr:

The following is a list of some of the speciﬁc measures that were attained about
services in modern facilities. These include general availability of services, including
hours per day and days per week the facility was open, as well as whether payments were
accepted in kind and whether there were sliding fee scales which changed with the
income of the user. We also know the total number of personnel and how many of each
particular type were in each center; these include medical doctors, student intern doctors,
registered nurses, nurses aides, midwives, etc. For treatment of each of six types of adult
illnesses, (pneumonia, bronchitis, gastroentiritis, TB, primary complex, and inﬂuenza),
we know what type of practitioner usually attends patients in the facility, the usual cost
for one visit, and whether drugs are available for this type of illness. For child-related
services, (prenatal care, child outpatients, immunization, and well-baby), we know
whether the facility even provides the service, usual cost and waiting times per visit, and

the type of practitioner who normally treats.

39

For drug availability, in the ﬁrst facility survey for each of the illnesses mentioned
above, the general question was asked whether drugs to treat this illness were available at
the facility. In the later survey, much more detailed information was obtained; here two
questions were asked, ﬁrst, whether the drug was usually available in the facility, and
second, what was the current stock of each of a list of drugs to treat various conditions.

In addition, both the usual and current availability of vaccines for children and pregnant
women (for BCG, DPT, polio, measles, and tetanus) and vitamin A and iron supplements
is given.

Information on traditional providers from the health personnel survey included the
individual’s education level, types of training received (including who conducted the
training session, the duration of the class, and when it was attended), whether infant
curative services are offered, attitudes about breastfeeding, contact with infant food
company representatives and the provider’s self perceptions of the inﬂuence of these reps
on their treatment practices.

While the Cebu survey provides much data on quality, ironically, the sheer
breadth of the data itself means that many of the variables are highly correlated, and that
degrees of freedom for estimation could become an issue. The data reduction method
employed was to construct indices which summarize different sets of aspects pertaining
to service quality (Peabody et al., 1994). This method was favored over other data
reduction methods, such as principal components or factor analysis, because the inﬂuence
on demand of speciﬁc sets of quality attributes can be directly assessed. The approach,
therefore, provides planners and policy makers with more useful information than an

aggregate quality index can.

40

In practice, the strategy is to create composite measures of different categories of
quality. Possible categories generally fall under one of three: structure, which refers to
the physical presence of resources and staff at the facility; process, which are the
practices followed by the health practitioners; and outcome, which refers to health
outcomes resulting from the care received (Donabedian, 1980, 198 8). As laid out above,
much of the data collected in the Cebu health facility instruments describe structural
attributes. While such structural measures cannot ensure higher quality care, they are
probably necessary for it. In addition, they can often be easily recognized by potential
users, so they may have a strong inﬂuence on demand (Garner, Thompson, and
Donaldson, 1990). Furthermore, the state of a facility's structural elements should reﬂect
resource availability, so that in an environment where resources are severely limited, as in
many developing countries, they may also serve as indicators of access to services
(Peabody, Rahman, Fox, and Gertler, 1994).

Health care attributes in our data which describe the physical presence of staff and
resources at the facility are staffing and drug supplies.20 The stafﬁng variables were
deﬁned as total number of personnel and the pr0portions of doctors and nurses relative to
total staff. Number of personnel may capture scale effects which could indicate shorter
waiting times and/or a wider variety of service availability. A higher proportion of
doctors to total staff may be perceived as providing better quality, while a higher

proportion of nurses may be viewed as providing services better than those of a midwife

 

2° Other studies have also included measures of facility infrastructure, such as
electricity, and plumbing, and equipment and supply availability, such as scales,
thermometers, stethoscopes, syringes, needles, bandages, etc. (Peabody, Rahman, Fox,
and Gertler, 1994). These types of data were not collected in the Cebu health facility
surveys, so are not among our quality indices.

41

or untrained practitioner; it may also be indicative of care that is more patient-oriented
and nurturing than a physician-attended visit. We experimented with numbers of doctors
and nurses also, but proportions were used because stafﬁng requirements at different
facilities vary according to level of care provided and size of facility, so actual numbers
of doctors and nurses cannot be directly compared meaningfully. We did not
experimented with other personnel types, such as dentists and pharmacists, because the
are not directly involved in providing basic curative services for infants.

Drug supplies are deﬁned as usual and current availability of ORT, intravenous
diarrhea treatments, child vaccines, and family planning methods. We used current, as
opposed to usual, availability. It was deemed to be a more objective indicator since it is
probably less subject to respondent bias; furthermore, current supplies could potentially
have been observed at the time of the survey by the interviewer, whereas usual supplies
could not without having many observations over time on the same facility.

As stated above, this detailed drug information was available only in the 1986
survey which made it more attractive for modeling health care demand. The year 1986 l-
was the time period immediately following the completion of the last household surveys.
To address the issue that the 1986 survey would not reﬂect the actual health care quality

situation faced by households from 1983 to 1986, perhaps because health facility

 

characteristics could have changed between the periods, means tests were performed on
several sets of the variables which were collected at both points in time explore

differences between years. Results indicate that the null hypothesis of no difference

42

between the two time periods could not be rejected for eight out of nine tests performed.“
Since characteristics measured in both periods were not signiﬁcantly different in 1986
than in 1982, this strengthened our conﬁdence in using just the 1986 data so that we
could have the advantage of including drug supplies in the models.22

We produced indices using most of the drug information collected in the survey. In
reference to diarrhea drugs, we expected these to be crucial determinants of demand since
they can have immediate inﬂuences on child health. Vaccines and the range of family
planning methods are obviously not related to child curative care, but may indicate an
orientation of the facility toward infant and maternal health services which could be
important to a mother in deciding whether this is where she takes her child. Furthermore,
mothers may be more likely to visit facilities with more vaccines and family planning
supplies for child curative care if they can access these supplementary services during the
child’s visit. We also tried including the availability of micronutrient supplements and
maternal tetanus vaccines. These, however, were not powerful predictors of curative care

demand, so were eventually dropped from the analysis.

 

2' The variables for which means tests were performed included doctor and nurse
ratios (discussed more below), number of child outpatients treated per week, and
outpatient waiting times. Separate tests were performed for public and private facilities.
The only test for which the null of equality between the two time periods could be
rejected was outpatient waiting time in public facilities, which rose from 2.5 to 6.8
minutes between the surveys. The test of the equality between the periods of the deﬂated
provided-reported user fee for private care could also not be rejected. A similar test was
not performed for public user fee because public facilities reported zero user fees in both
years.

22 It could be argued that drug supply characteristics can change more quickly than
some other health care attributes, and therefore, they could have differed between the two
time periods. However, we have no data with which to test this hypothesis.

43

Other more process-oriented variables we experimented with included an indicator
of whether the establishment treats child outpatients at all, average waiting times for child
services, and the number of child outpatients treated per week at the facility. The latter
two variables were not kept because their predictive power was low.

Regarding the traditional practitioner characteristics, while provider contact with
infant food company representatives and attitudes concerning breastfeeding could
inﬂuence the advice they give concerning child feeding practices, these were not viewed
as being crucial determinants of infant curative care choice, so the education and training

were chosen as the descriptors of traditional practitioner quality.

beam:

Although price information was collected in the health facility surveys, this data
was not used in the analysis for several reasons. First, there was no provider-reported fee
data collected for traditional practitioners. Second, there are many cases of missing
values for price at facilities; and third, for public services there are no official fees
reported, so there were only zeros reported in the data. Often in empirical work, if price
data is not good or not available, expenditure data is used; the approach is to compute
average prices per unit of the good within geographic sampling clusters and use these as
measures of price. Unfortunately, however, expenditures on child curative care were not
collected in the Cebu study. Data was collected, however, on expenditures for prenatal
care visits. While this is a different type of service, it is the most complete source of user
fee data in the survey. Using this variable we attempted to construct hedonic prices for

each individual; the extent of information available in the survey on health care market

44

conditions was not sufﬁcient, however, to achieve identiﬁcation of a hedonic price. This
being the case, barangay median deﬂated prices were constructed for each provider type.
We can compare these prices to those reported by providers in the facility survey where
ofﬁcial user fees charged for prenatal care and for child outpatients were given (shown in
Table 4.1). Public providers reported charging zero for both services; the average
barangay median values for public care is 0.54 pesos (1980). Private provider fees for a
prenatal and child outpatient visit were 3.25 and 8.75 pesos, respectively, while the
barangay median prenatal fee constructed from the household expenditure data is 4.83
pesos. Barangay median user fee for a prenatal visit to a traditional provider was an
average of 1.06 pesos.

Regarding distance and travel costs for care, information was collected in the
baseline survey on travel times and travel costs to providers used by mothers for prenatal
care. This information could have been used in the choice analysis, but in a strict sense it
is endogenous because distance is a function of health care choices made by mothers.
Another alternative would have been to use barangay average or median values of these
travel variables. Supplementary data was available”, however, on distances from each
household to each of the modern health facilities used by sample mothers.24 This enabled
us to have distance from each household to the nearest of each type of health facility,

which is an exogenous measure because it is not a choice variable.25

 

23 The author would like to thank David Hotchkiss for providing this data.

2“ Strictly speaking, the distance is that between each health facility and 49
geographic points in the area with represent household clusters.

25 This assumes that households have not migrated to this location because of
attributes of health facilities in the area. It also assumes there is no purposive placement
of facilities in areas with high demand due to high morbidiy or high incomes.

45

Distance data from each household to the nearest traditional provider was not
available in this same data (probably because many work in less formal settings than
modern providers, making them more difficult to locate). For these providers,
community-level information was used on whether there was a traditional provider in the
community; if there was, distance was set to zero; if there was not, distance to the closest

available one outside the community was used.

DEFINITION OF CURA TIVE HEALTH CARE CHOICES

The manner in which the different facilities choices are deﬁned is critical to the
analysis. The household survey originally included responses for ten different types of
facilities that were used infant outpatient curative care. Retention of all the original ten
facility types would have resulted in a discrete choice model with small cell sizes for
several of the outcomes; this would have been particularly problematic for analysis of
responses by different subgroups.

Given that one of our primary interests in this research is to investigate household
responses to public health care policies, we deﬁnitely want to separate public care from
other types of care. Furthermore, because western and traditional providers vary in their
treatment approaches, we may also wish to separate traditional care from “modern” health
care services.

The next stage of disaggregation to consider is a breakdown within public and
private services by facility level. In many studies, hospitals and clinics are separate
choices. In the Cebu data, however, child curative visits to hospitals, both and public and

private, were extremely infrequent. Only 2.5 and 1.1 percent of the total 31,000 person-

46

observations were for a public and private hospital visit, respectively. The options were
to either drop the hospital observations or to aggregate hospitals and clinics into a single
category for each of the modern sectors. The latter strategy was followed even though
some facility attributes differed, as presented in Table 4.2. Quality variables are
displayed for modern services, broken down by clinics and hospitals; public clinics are
divided into levels as well, barangay health stations being the smallest and most basic. In
the public sector, user fee, nurse ratio, and vaccine and family planning indices are
somewhat similar in clinics and hospitals; in the private sector, user fee and family
planning index are somewhat similar. The ﬁnal health care alternative set used consists

of public, private, and traditional health providers.

c . . F . i .

Quality attributes in the data analysis ﬁles represent the attributes of the closest
facility of each type to each household; for modern facilities this matching was
accomplished using the distance data. For traditional providers distance to households
was not available, so we constructed barangay-level median values for each of the

traditional provider characteristics in the health personnel survey.26

PRICES AND QUALITY OF THE RESPECTIVE FACILITY TYPES

 

2‘5 Some traditional providers used by the sample households were outside the
sample barangays. In order to use data on all these providers, we assigned each
traditional provider outside sample barangays to the closest sampled barangay. We then
took the median value for each quality aspect as the barangay-level value used in the
analysis. There were three barangays for which no traditional providers were
interviewed; for these we used the value of the closest barangay of same urban/rural type.

47

As seen in Table 4.2, distance to the nearest provider and user fees are greatest for
private facilities. Public services are closer and charge either no fee or an extremely low
one; traditional providers are the most accessible geographically, and while their fees are
low, they are higher than those charged by modern public establishments. Differences in
services at the three types are also wide: public and traditional facilities all provided
child outpatient care; however, ten percent of the private facilities did not. Private
facilities have more staff and higher ratios of doctors to total staff relative to public
facilities; nurse ratios are similar for public and private. A very large percentage of
public centers had ORT supplies on hand, but very few had more complex diarrhea
treatments. On the other hand, less than half of all private establishments had ORT
available, but quite a large proportion had more complex diarrhea treatments, such as
intravenous solutions, in stock. Vaccines and family planning supplies are slightly more
available at public centers. For traditional providers, there is only one staff member in
the 'facility'; by deﬁnition doctor and nurse ratios are zero; modern drug supplies are also
set to zero. Service quality for these providers is measured by education level and

whether any formal health training had been received.

DESCRIPTION OF PA ITERNS OF HEALTH CARE UTILIZATION

Health care utilization patterns are given in Table 4.3. For the entire sample over
all rounds, a curative health care visit in the two months prior to each survey occurred
forty-nine percent of the time. Among the three types of visits, traditional practitioners

are used most frequently, followed by private, and then public services.

48

By log of household asset value, overall visits increase gradually with values up
to the ﬁrst tercile of the distribution, and then level off. Health care visits are only
slightly greater among high asset households; however, the composition of visits differs
by asset status. Utilization of public care is nearly constant up through the ﬁrst tercile
value where it begins to diminish; private visits, alternatively, rise slightly through the
second tercile and then increase dramatically. Traditional care is used at about equal
levels for households in the ﬁrst two terciles and much less by the high tercile
households.

By maternal education, visits increase with the ﬁrst eight years of mother
education and more slowly thereaﬁer. Public use increases slightly up to the sixth year of
schooling (primary school completion), then rises very sharply between six and eleven
years, and then decreases; composition of the visits is very different for the low and high
education group. Use of private care rises with maternal education, especially after grade
ten, while visits to traditional practitioners drop if the mother has more than a primary
school education.

By child age, demand rises up to the child’s six or seven month and then falls. Use
of modern care, especially private, is much higher for children in the ﬁrst six to eight
months of life relative to older infants: approximately forty percent of all 0—6 month old
visits are at modern facilities versus only twenty-ﬁve percent for those in the 7-24 month
old group.

Trends by calendar month indicate that utilization of private care is lowest in the

wet season, June-September (a high diarrhea period), and greatest in the dry season,

49

February to May (a high FRI period). Traditional visits peak in the middle of the dry-

and wet-seasons. Demand for public services appears to be lowest during the dry season.

COMMUNI T Y-LE VEL NON-HEALT H FACILITY VARIABLES

Construction of the food price variables used for the analysis was a multi-step
process. Market food prices were measured at the community-level at ten different
periods of time during survey. We ﬁrst assessed completeness of price and unit data for
each item, in each community, during each round. Within broad food groups, we
aggregated brands so that a range of prices that reﬂected different levels of within-food
group quality would be present. The major food items of interest for our purposes were
infant formula, corn, which is the major staple commodity, and cooking oil. Foods that
came in different size containers were transformed to a common unit (e.g, ml, gram, etc.),
so that unit prices could be constructed for each community and for each round. These
unit prices were then deﬂated using month—year CPI information (Philippine Statistical
Yearbook, 1988). Finally, for communities with missing values for a particular item in a
particular month, prices from neighboring barangays from the same time period were
used for purposes of imputation. Food prices are generally lowest between the dry and
rainy seasons, begin to rise in the dry season, and peak during the wet season.27

Other non-facility community variables describe the health and sanitation

infrastructure, the disease environment, and other non-health physical infrastructure.

 

27 It is probable that these prices follow the agricultural calendar where the interim
season is when the harvest occurs; the dry season is when food stocks start begin
diminish and energy required to prepare ﬁelds for planting is high; the wet season is
when cultivation begins and food stocks are at their lowest.

50

These are described in Table 4.5 with summary statistics in Table 4.6; they include the
proportion of households in each community having, respectively, piped or pumped water
to their house, a reﬁigerator, a modern toilet, and sanitary garbage disposal methods;
whether the community has frequent water shortages (a common problem in the area), the
availability of a bank, or improved roads. Also included are community elevation, which
helps capture infrastructure and temperature, present and lagged values of rainfall levels

and present and lagged values of infant formula, corn, and cooking oil.

INDIVIDUAL- AND HOUSEHOLD-LEVEL VARIABLES

Individual- and household-level variables included in the analysis, described in
Table 4.5 with summary statistics in Table 4.6, include age and sex of the index child,
mother’s father’s education and age, mother's height”, and a dummy indicator for
whether each parent was physically absent from the household during the entire ﬁrst two
years of the child’s life.29 Furthermore, because household characteristics should include
measures of accumulated human capital of parents, maternal height is used to capture
some of these attributes which are not picked up by her education.30 We also include
household structure variables which may reﬂect household-level time and resource
constraints relevant for child health care utilization and for child health production. For

instance, the presence of other infants in the household may contribute to index child

 

28 Unfortunately, father's height is not available in these data.

29 If absent for all twelve survey rounds, the person’s characteristics are set to zero in
the regressions. If the person was only temporarily absent, her or his individual
demographic information is retained for every round.

3° Even with these variables, however, parental human capital is probably still not
measured completely.

51

illness and/or resource crowding; the presence of the elderly who require care could make
for greater household time constraints; additional adult females in the household,
however, may have positive effects for child health inputs if this group has stronger
preferences for child health.31 Maternal age, education, and height are also used to
capture exogenous underlying determinants of the mother’s wage rate.32 Deﬂated value
of household assets at the baseline survey are used to capture household resources. The
asset variable was constructed from detailed information on ownership and values of
items in the following categories: houses, land, vehicles, livestock, agricultural and
business equipment, fumiture, household appliances, and kitchen equipment.33

Several of the household variables are entered as linear splines; these include

child age, mother and husband education, mother height, and household asset values.

 

3' It is arguable whether household composition variables should be treated as
exogenous in a model of health care demand; assuming that they are exogenous implies
that fertility decisions and other household composition changes are exogenous for child
health care demand. Including them does not change the signs or signiﬁcance of the
other household variables (as discussed in Section 6 and shown in Table 6.5), so they are
retained in this study.

3’2 Wages were investigated as explanatory variables but are not used in the estimation
for several reasons. First, in the household survey, many individuals, especially women,
report not having income from wages. Second, forty percent of the female wage
observations come from the baseline survey when most of the women were in their last
trimester of pregnancy. This value of time is probably not what it would be under normal
conditions. Third, many of the wages are classiﬁed as "self-employment" wages. These
values should not be used to infer market wages because of the difﬁculty of
distinguishing net income from an enterprise vs. returns to entrepreneurship, risk-taking
and capital investment. Fourth, barangay level data on wages from the community
surveys was sparse and showed little variation. Finally, the community data do not
contain sufficient information on local labor market demands, sectoral composition,
unemployment rates, etc., to use as exclusion restrictions for identiﬁcation of wages in
the health care demand equations.

33 Price shocks were large during this period, particularly during 1983 and 1984.
Monthly price data collected for the region by the National Census and Statistical Ofﬁce
of the Philippines is used to deﬂate nominal values.

52

Spline transformations provide a way to assess the relationship between an explanatory
variable and an outcome of interest semi-parametrically. The variable is divided into
piecewise linear segments, and the coefﬁcient on each interval represents the slope for
that interval. For example, the coefﬁcient on the ﬁrst segment of the child age variable
gives the effect of an additional month of age up to the sixth; the second segment gives
the effect of an additional month of age after the sixth. For each regressor, the hypothesis

that the slopes of the adjacent segments were equal was rejected in each case.

53

Section 5

EMPIRICAL MODEL

INTRODUCTION

The demand for curative outpatient services for children is the outcome examined
in this study. It is deﬁned as the initial facility type chosen for a consultation if in fact the
child had any curative visit during the the two months preceding each longitudinal
survey. This is as a discrete outcome since we are not modeling number of visits or
expenditure on visits. For purposes of the analysis, the relevant service categories
available are modern public, modern private, traditional private, or no provider. As
discussed previously, these options differ substantially in terms of price and quality of
service.

The demand for a particular alternative is the probability that it yields the highest
utility among those available. In a discrete modeling framework, this probability is
interpreted as the demand function. The functional form of the demand function depends
on the functional form of the conditional utility function and the distribution of the
stochastic terms. We assume utility is linear in health and consumption, which implies

the conditional utility function shown in (7) is now

(9) Uj = a” Hj (Cj, Fj*; S, M, E, U) + an [Yj — BjCj - wTJ-Cj ~ pFFJ-l‘] + ej

where F j“ is the optimal choice of other health inputs given health care choice j, and (—:j is

a zero mean random disturbance term with ﬁnite variance and is uncorrelated across

54

alternatives and individuals. Because the data do not provide sufﬁcient information to
construct a household income variable, we cannot directly model utility derived from
consumption. We use the value of the household's total assets as a proxy for household
income; assets are reﬂective of the household's long-run resources, so are highly
correlated with income and consumption; moreover, liquid assets play an important role
in consumption smoothing.

Parents are assumed to make health care choices based on the comparison of the
indirect utility functions for each type of health care alternative available, including that
of no treatment. In practice, speciﬁcation of the demand across alternatives is based on
the difference in the utility of each market care alternative from that of self care. Under
the assumption that their are no user fees or access costs for self-care, the conditional

utility function for self-care alternative is

(10) Us = als Hs(C39 125*; 89 Ms E, U) + “25 [Y5 ' pFFx*] + 65

So the difference in utility between each market health care option and self-care is

(11) Uj " US : an Hj (C F‘*. S, M, E, U) -(X.ISHS (CS9 FS*; S, M, E, U) +112ij 'asts "

l’l’

[aszFFj* -a23pFFs*] 'a2j[BjCj ' WTjCj] + Ej -€s fOI'j is

This is the structural demand equation. The Option giving the highest (UJ- - Us) is the one

chosen.

55

Substituting out for the reduced-form determinants of H, Y, F, and C, we obtain
the reduced-form indirect conditional utility ﬁmction for each alternative. These
equations express the conditional utilities in terms of income, prices, and other reduced-

form determinants. This leads to the estimated speciﬁcation

(12) Vj = 1301' +9” S +sz M +1331 A +l34j E +p5j Bj +l36j WTj +137j Qj +p8j PF +B9j U +€j

forj=1toJ

where j is the type of health care chosen; S is a vector of individual child characteristics;
M is a vector of household characteristics; A is the value of household assets; E is a
vector of community health characteristics; Bj is the user fee for health care choice j; w is
the wage rate and Tj is the time incurred to obtain health care from choice j. The variable
v captures individual child and household unobservables and it includes elements such as
innate healthiness of the child and household-level heterogeneity in health technology
and preferences.34

The error term is assumed to have a Gumbel distribution, leading to the

multinomial logit speciﬁcation.

 

3" These unobservables are dealt with ultimately by employing robust standard errors
which are corrected for repeated observations on individual mother-child pairs in the
empirical work. Panel data methods are not used to address unobservables for two
reasons: ﬁrst, many of the variables of interest for this study, particularly certain health
facility quality information, are observed only once in this survey; second, panel data
methods for unordered polytomous limited dependent variables, such as the facility
choice model in this study, are not well developed.

56

(13) PJ. = exp(Vj)/Z [exp(Vk)] for k = l to J

For our model this would amount to estimating

(13') Pj = LDLQOjiﬁrjAS—ﬂiszu ﬂjﬂﬁj—BjiﬂéjﬂjiﬁﬁﬂjiﬁﬁgFi-ﬂﬁ U_i‘§j)
ij=reXPmOj +13” 3 +sz M +B3j A +B4j E +133 Bj +p6j WTj +p7j Qj +p8j PF +13ng +6)

for each of the four health care choices. If we simplify the notation and refer to the

estimated set of coefﬁcients from choice j as B“) and the set of regressors as X, then we

can write (13') as

(13") Pj = exp(XB‘j’)/z [exp(XB(‘")] for k = l to J

One feature of the multinomial logit (MNL) model is that it is unidentiﬁed in the sense
that there is more than one solution to 13“" 13(2), [3(3), and B“) that leads to the same
probabilities for j=1, j=2, j=3, and j=4. To identify the model, one of the [30’s is

arbitrarily set to zero; it does not matter which one. For instance, if we arbitrarily set the

-_.- "—_l

self-care alternative [5“) to zero, the remaining coefﬁcients, [3“), [3(2), and 13‘” would

measure the change relative to the self-care option. In the modeling, this is the category

 

we set to zero for ease of interpretation of the coefﬁcients. This yields the following
probabilities

P0) = exv(l3‘” X) / exr9(13"’ X) + exp(l3"’X) + eXP(B‘3’X) + 1

ND =eXp(ll‘2’ X) / eXp(13“’ X) + eKP(13‘2’X) + eXP(B‘3’X) + 1

P(3) =eXp(l3"’ X) / eXP(B“’ X) + eXp(B‘2’X) + e><13(15""X) + 1

57

P(4) = 1 / eXP(l5"’ X) + eXp(13‘2’x) + 6XP(13"’X) + 1

and the following relative probabilities
P(1)/P(4) = eXP(B"’ X)
I’(2)/P(4) = 6XP(B‘2’ X)

P(3)/W4) = eX13(13‘3’X)

SPECIFICATION ISSUES
ALTERNA T IVE-SPECIFIC INCOME COEFFICIENTS

Note the altemative-speciﬁc subscript on income in our baseline speciﬁcation in
(12). Gertler and van der Gaag (1990) argue that allowing separate, altemative-speciﬁc
income coefﬁcients in a discrete demand model of health care choice would violate
rational choice axioms. The reasoning is that if two health care alternatives are available
for same price and each gives the same improvement in health, then the income
remaining after paying for health care is the same in either case. Therefore, if utility is
affected only by health and consumption opportunities, individuals should be indifferent
between the choices. One rationale for not constraining income effects to equality across
health care types, argued by McFadden (1981), is that tastes for each alternative may
depend on individual characteristics that are correlates of income, such as assets, and
historical wages rates and income levels, and because income is typically measured only

by its correlates, different income coefﬁcients are justiﬁed.

58

FLEXIBLE HEALTH CARE PARAMETERS

Another unique feature our baseline model shown in (12) is that parameters on
each of the facility attributes is allowed to vary by type of care. This approach is more
ﬂexible than that used by most other health care demand studies. Given the wide
variation in the nature of the service types, e. g., personnel levels and training, drug
availability, and inevitably other unmeasured aspects of service, one can make a strong
argument that care from the different segments of the health care market can reasonably
be considered to be different goods. In a recent paper which compares various
assumptions underlying previous discrete choice models of health care demand, Dow
(1995b) concludes that constraining price and quality coefﬁcients to be equal across
health care alternatives is the most strongly rejected of all. He also shows that imposing
this assumption can have large effects on elasticities, which is important given the policy
focus of responses to user fees. Another weakness of the constrained approach is it does
not allow different sets of health care characteristics to impact the probability of visiting
different types of providers. Forcing divergent ﬂavors of health care to be inﬂuenced by
the same set of attributes, and imposing the restriction that each of these attributes has the
same effect on every kind of service, is unrealistic at best and could be highly misleading.
It puts unnecessary limits on the insights which can be drawn concerning how prices and
quality affect demand for care from different segments of the health care market.

In the empirical work we explore how imposing this constraint affects our results.
Equation (14) gives a "constrained" version of the conditional indirect utility ﬁmction,
where all health care choices are forced to have the same set of attributes and the

coefﬁcients on these variables are forced to be equal across facility types. Note that the

59

 

estimated coefﬁcients for the health facility attributes,B,-, T], and Q], no longer carry the

subscript j.

(14) V,“ = Raj +Prj S +BZj M +133j A +p4j E +l35 Bj +136 WTj +137 Qj +l33 PF +139; U +51“

This model is expected to yield very different results from the baseline ﬂexible

speciﬁcation.

COMPLETE CROSS-EF F ECT S VERSION

In both empirical speciﬁcations discussed above, the only provider attributes that
enter each indirect utility function are own-characteristics. Those of the other alternatives
enter the model when the decision-maker compares expected utility from each respective
provider and chooses the one yielding the highest Vj. Another approach, shown in (15),
is to allow characteristics of substitute providers to enter directly into each Vj. Allowing
attributes of all health care choices to enter the utility of each choice j could be
considered more analogous to a typical demand function.” An alternative interpretation
is that agents employ forward-looking behavior (Dow, 1995b); for example, a person may
choose to visit a public facility today depending on the price of a private (or traditional)
visit tomorrow.36 Note that the facility coefﬁcients in this "full-effect" model now have

"jk" subscripts:

 

35 See for example the discussion by Dor and van der Gaag (1987).
36 This assumes that prices tomorrow will be the same as prices today. Note that
dynamic objective and budget constraints are not explicitly dealt with here.

60

(15) Vj =l30j +Blj S +132j M +B3j A +B4j E +2135jk Bj +Zp6jk WTj +2571} Qj +st Pr +1391“ +€j

This model allows the exploration of cross-price and cross-quality effects which are not
allowed in the baseline ﬂexible model. One possible issue that may arise with this
speciﬁcation is that certain facility attributes, such as time prices for care, could be highly

correlated across facility types, especially public and private modern care.

ECONOMETRIC METHODS

As discussed above, the model is estimated using multinomial logit; the choice

probabilities are laid out in equations (13) and (13'). The marginal effects for the model

are
(16) an / 6xji = bj, Pj (l-Pj) (own)

de / 6xki = bki Pj Pk (cross)
An important property of the multinomial logit model is the independence of irrelevant l—

alternatives (IIA), which states that the odds of facility type i being chosen over facility

type k is independent of the availability of alternatives other than i and k. This implies

that an additional alternative could be added to the model without changing the odds

 

'71"—

ratios of the original alternatives. It should be noted, however, that if any of the
alternatives are similar, this may be an unreasonable restriction to place on household

behavior. The model produces equal cross-substitution between any pair of alternatives

61

in the presence or absence of other alternatives because the error terms across alternatives
are assumed to have zero correlation.”

A more general discrete choice model which is able to accommodate different
structures of error term correlation is the nested multinomial logit model (McFadden,
1981). The NMNL allows for correlation across subgroups of alternatives and, therefore,
nonconstant cross-price elasticities; it allows grouping of alternatives which are closer
substitutes so that cross-price elasticities are more elastic within groups than across
groups (Gertler and van der Gaag, 1990).

With this approach alternatives are partitioned into groups that are similar to one
another. One possibility for our four health care choices would be to combine the market
health alternatives into one group, with the reasoning that they are more similar to one
another than each is to the self-care option. If we consider these to be different “levels”
of a choice tree, the choice to visit a facility or not is in one level, and what type of

facility to choose is in another:

Visit No Visit (Limb: L)
| I |

Public Private Traditional (Branch = C)
Another more disaggregated version of the decision tree would be to group the modern

 

 

public and private provider types together on the grounds that they differ from traditional

practitioners.

 

37 In version of the model where cross-effects are introduced the Independence of
Irrelevant Alternatives (IIA) assumption is relaxed.

62

Visit No Visit (Limb= L)
l 1
Modern Traditional (Branch = C)

| |
Public Private (Twig = T)

 

 

Still other possibilities exist, such as separating hospital- from clinic-level care within
each type of modern alternative. As explained in the data description, however, there
were only a very small number of hospital visitors in this sample (four percent of the
observations), so this was not a feasible method of dividing modern visit types for this
study. The two decision trees above are the ones experimented with in the empirical
work.

For the two-level model, the probability of choosing any particular facility type
can be expressed as the product of a conditional probability and a marginal probability:38

(15) P6,, =exp(Vc)/Z[exp(Vn)] (Condit'l)

for n=1 to C facility type branches within the visit limb r?

and P, = P(visit, no visit) = eXPWz + Tili) Q: [CXPlvm + 1:mlmll (Marginal)
for m=1 to L limbs, where I=log[£exp(Vn)]

and P0, = Pct. * P, (Product)

where Pc is the probability of each of the C conditional choices, i.e., the probability of
choosing each of the market health care alternatives c within the "visit" limb i. P, is the

probability of choosing each limb i, i.e., the probabilities of choosing a visit and no visit.

 

3“ See Appendix B for the derivations of (1 5)-(1 7).

63

The variable I is the logarithm of the denominator of the conditional choice within limb .6.
This what is known as the "inclusive value" for the branches c in limb t. The parameter,
1:, of the inclusive value provides a basis for identifying the behavioral relationship
between choice at each level of the tree. If the estimated parameter, 1:, is equal to one, the
two levels can be collapsed to one level. For this model, it would imply that the two-
level NMNL model would collapse to the single-level multinomial logit structure.

In estimating the two—level tree, we allow the health care characteristics to enter
only in the facility-choice decision, while the individual, household, and non-facility
community variables enter both decision levels, implying that they inﬂuence both
whether a visit occurs, and conditional on a visit, the type of facility chosen. Allowing
the demographic and community variables to enter more than one level of the decision
tree is an unusual innovation to the NMNL model. Most models allow any particular
regressor to enter in only one level, which may not always reﬂect the decision-making
process well.

The unconditional marginal effects for the facility variables were derived
analytically for this specialized NMNL speciﬁcation (see Appendix B) and are deﬁned
as:

(16) ape / axci = bci Pcll PI [(1- Pcll)+(1-Pl) T Pelt] (OWN)

apn / axci = bci Pcll P! Pn|l [1(1'1)! ) '1):l (CI'OSS)

64

Unconditional marginal effects for individual, household, and non-facility community
variables in this specialized NMNL model were also derived analytically and are given
by:

(17) ape / axi : Pelt Pl [bci (1" Pcl') + (a! + 1'- Pcll bci )(1'P1)]

The nested logit model can be estimated sequentially or simultaneously using
maximum likelihood methods. For sequential estimation, the conditional facility choice
is estimated as a MNL model, then the inclusive value is calculated for the conditional
choice set and included as an regressor in the marginal choice of whether to have a
market care visit or not, which in this case is estimated as a logit model. The parameters
estimates for the facility characteristics are efﬁcient for the subset of the health care users
since these variables appear only in the facility-choice level of the tree. Those which also
appear in the market vs self-care level, i.e., all the non-health facility variables, are
consistent but not fully efﬁcient due to the use of "estimates" in obtaining the inclusive
value parameter (Amemiya, 1978). Since this estimate is used in accepting or rejecting
the nesting structure, obtaining an efﬁcient estimate is important. McFadden (1981)
provides a correction procedure to adjust the standard errors of parameters in the marginal

choice model; it is extremely complicated, however, even for a two-level model

 

(Hensher, 1986).

An alternative route is to estimate the model simultaneously by maximum
likelihood methods and obtain fully efﬁcient estimates of the parameters. This is the
preferred approach; in practice, however, obtaining FIML estimates for this model proved

difﬁcult for two related reasons. First, the model contains a large number of non-health

65

facility explanatory variables whose effects vary by health care alternative. When
estimating a FIML nested logit model, any variable that does not vary by alternative must
be interacted with a choice-speciﬁc constant, otherwise it drops out of the estimation.
This increases the number of parameters for this set of regressors from S to S*(J-1); in
this model it means the number of non-health facility parameters was essentially tripled
from 62 to 186. Second is that estimated parameters on the health facility variables are
not constrained to equality. This is one of the primary features of the model, and means
that instead of C of these we have approximately C*(J-1), increasing the model's size
even more.

The method ultimately employed was to estimate the model sequentially and use
the bootstrap sampling method to correct the standard errors for the fact that the inclusive
vales are estimated.39 This amounts to estimating the full decision tree many times over,
with N observations being drawn each time with replacement from the N observations.

In this random drawing, some of the original observations will appear once, some more
than once, and some not at all. At each pass (called a replication), the estimator is applied
to the data and the resulting parameter estimates are saved as a data set. Using the
collection of the parameter sets resulting from these replications, one can calculate the

standard deviation of each statistic, which is an estimate of its standard error (StataCorp,

 

1997):“O

 

39 Bootstrap methods are detailed in Efron (1982) and Efron and Tibshirani (1986).
4° Although the average of the bootstrapped statistic is used in the calculation of
the standard deviation, it is not used as the estimated value of the statistic itself; the point
estimate is the original observed statistic computed using the original N observations

(StataCorp, 1997).

66

For this particular study, we have repeated observations on mother-child pairs.
Referring to each of these pairs as a cluster, this intra-cluster correlation will result in
artiﬁcially low standard errors if the repeated sampling is not taken into account when
calculating the standard errors. We therefore identify these as clusters for resarnpling so
that the sample drawn during each replication is a bootstrap sample of clusters.

A potential weakness of the sequential approach is that the lower levels of the
model are estimated using observations only on those individuals who (1) had those
particular alternatives in their feasible choice set, and (2) actually chose one these
options. The ﬁrst item is not problematic in this study because all persons had each of the
alternatives available to them. The second issue is a concern, however, because persons
choosing market health care at any point in time may be a select group and different in
both observable and unobservable ways from those who do not; using only the subsample
of individuals who chose to use market care in any particular period may result in biased
parameter estimates (Hensher, 1986). For our study, only half of the sample observations
had a health care visit; therefore, with the sequential NMNL approach only half the
observations can be used to estimate the determinants of facility choice conditional on

there being a health care visit.

67

Section 6

RESULTS FOR THE BASELINE POOLED MODEL

W

Results for the binary outcome of demand for any type of health care visit are
presented in Table 6.1; these are logit estimates with standard errors corrected for the
intra-individual correlation in the data which arises from the presence of repeated
observations on the same mother-child pairs over the twelve rounds of the survey.

The health care variables, whose effects are allowed to vary by type of health
facility they describe, were entered in successive steps in the regressions for the purpose
of investigating how the impacts of health care attributes on demand change when
moving from the commonly speciﬁed distance-only model to one that also includes user

fees, and then to one that includes distance, user fees, and health care quality.

INDIVIDUAL AND HOUSEHOLD INFLUENCES

Across each speciﬁcation, the inﬂuences of individual and household
characteristics are virtually identical and will therefore be discussed ﬁrst. The results,
presented in Table 6.1,41 reveal that all else being equal, male children are more likely to

be taken for a curative care visit. Demand rises with child age up to six months and

 

4' The results for the health care attributes from this same regression are presented
in the ﬁrst column of Table 6.2. The results for the non-health care community effects
from this regression are presented in Table C1 of Appendix C.

68

declines thereafter, as was seen in the bivariate graphs."2 These ﬁndings highlight the
unconditional nature of the estimates. Boys are more likely to be taken for a health care
visit despite the fact that boys' illness rates do not differ statistically from girls' over the
entire age range of the data (2 to 24 months).43 By child age, demand for health care and
reported illness levels (using 24-hour recalls of detailed symptoms) increase up to the
child's sixth month; however, demand falls after the child's sixth month, while reported
morbidity remains relatively constant instead of declining.“

Maternal education and height increase the unconditional demand for services.
Recall that the unconditional estimates are interpreted as the sum of the short-run
elasticity of curative care, plus the elasticity of morbidity with respect to the covariate of

interest. We expect maternal education to increase the former and reduce the latter

 

42 As explained in the data discussion of Section 4, the spline variables are
constructed so that when, used in estimation, the coefﬁcients represent the slope for that
interval; i.e., the variable is divided into two piecewise linear segments. For example, the
coefﬁcient on the ﬁrst segment of the child age variable gives the effect of an additional
month of age up to the sixth; the second segment gives the effect of an additional month
of age after the sixth.

An alternative method of estimation using spline variables is to have the second
segment measure the change in the slope from the preceding interval. That IS NOT the
method used here.

‘3 See footnote 6 in Section 1 for a description of illnesses levels by child age and
sex.

4“ Alderman and Gertler (1997) ﬁnd in rural Pakistan, however, that conditional on
an illness being reported, the probability that no health care is used (i.e., only self-care is
chosen) for a daughter is only 0.015 percent higher than for a son.

An important issue, though, related to the conditional nature of their sample--only
children who are reported to have been ill are included. If there promale bias for
investing in child health, this could lead to under reporting of girls' illnesses, so that only
girls who are very sick will appear in the sample. If this holds, it is possible that the pro-
male bias could appear in an unconditional version of their model.

69

effect;"5 therefore, the net positive effect of education implies that its inﬂuence on
curative demand is strong enough to outweigh its morbidity reduction effect. Maternal
age reduces the probability of a health care visit; perhaps controlling for education, older
mothers have more experience in rearing children and are better at producing healthier
kids who in turn need less care; on the other hand, older mothers may have preferences
for not using formal health care, possibly because they are less familiar with the service.

Household asset values up to the ﬁrst tercile signiﬁcantly raise the chances of a
visit; assets beyond the ﬁrst tercile do not have signiﬁcant effects.46 It is conceivable that
after the ﬁrst tercile value, most households can afford at least some type of health care
and beyond this level, additional household resources probably inﬂuence more the type of
care chosen; additional household resources could also reduce demand for curative care if
higher resources contribute to better health status.

The presence of additional children, particularly males, in the household who are
younger than the index child reduces the probability of a curative visit for the index child.
This result could arise for a variety of reasons. Additional children could tighten
household income constraints, thus reducing number of visits for all children. However,
because the effect is greater for additional male than female infants, this is probably not

just a simple resources crowding story, but may indicate a preference for newborn males.

 

‘5 Bivariate graphs reveal that the use of curative care increases at a relatively
constant rate with maternal education; reported morbidity levels, on the other hand,
remain constant up until the third education (around 9.5 years of education), when they
fall. We expect actual morbidity levels, however, to decrease more or less monotonically
with maternal education.

‘6 Bivariate relationships reveal that visits increase with log asset values up to the
ﬁrst tercile and then level off; illness rates are relatively constant with log asset values in
the ﬁrst two terciles, and then decline with the third tercile of log asset value.

70

Furthermore, given that reported illness levels increase sharply during the ﬁrst few
months of life and boys' rates rise slightly faster than girls' during this period, parents
may therefore be acting in a manner they see as rational by investing ﬁrst in the health
care of the youngest children, especially those who are male, because they may be (or at
least perceived to be) the most vulnerable to illness.

Elderly men in the household also have a signiﬁcant negative effect on the chances
of a health care visit for the index child. Although it is possible that the presence of these
men could reduce demand by lowering child illness, it is probably more likely that they
impose additional time burdens on those who provide care to the very old and young in
the household; they probably also compete for resources, reducing the amount remaining
for child health investments.

In sum then, health care visits for index children are increased by their being male
and younger than six months of age, and by maternal human capital and household
resource levels. Maternal age as well as the presence of other very young children and

very old men in the household reduce the demand for child curative care.

HEALTH FACILITY INFLUENCES

Turning to the inﬂuences of health care attributes, displayed in Table 6.2, in the
initial model where distances to the nearest facility of each type are the sole health care
attributes in the model, remoteness strongly reduces the chances of a visit: kilometers to
the closest of each of the three provider types have negative and signiﬁcant impacts on

the demand for care.

71

With the inclusion of user fees, the only noticeable distance effect remaining is that
to the nearest private facility, and its inﬂuence is strengthened both in magnitude and
signiﬁcance.47 Among the user fees, public price appears to signiﬁcantly increase the
demand for care; traditional prices have smaller, but still positive and signiﬁcant effects
on demand. We would not expect these types of results unless the services of higher
priced facilities are in greater demand because better services are offered.

When the quality data is added to the model“, none of the distance results change
from the distance-user fee version. Public price becomes virtually insigniﬁcant and its
coefﬁcient is reduced slightly, although it remains positive; we surmise that this
remaining positive impact stems from still omitted service attributes. The private and
traditional price effects lose any signiﬁcance they had before quality was controlled for.

For provider quality, each attribute that was statistically signiﬁcant had a positive
effect on utilization. The number of personnel, range of vaccines offered, and availability
of ORT stocks at public clinics have strong positive inﬂuences on demand for care.
Private staff levels and doctor and nurse ratios increase demand for services. None of the

traditional practitioner results were powerful predictors of a health care visit.

EFFECTS OF EXCLUDING NON-FACILITY COMMUNITY CONTROLS

 

"7 Given signs of user fee variables, we would have expected the distance effects to
have gone in the opposite direction. It is possible that correlations between distance and
user fee within each facility type be affecting this. Distance-fee correlations:
public: .0068; private: -.2488; traditional: .1470.

‘3 This is the regression for which individual and household effects are presented in
Table 6.1, and community effects are presented in Table C1 of Appendix C.

72

An important issue in measuring how provider attributes affect demand is the
existence of many potential community inﬂuences, the difﬁculty of measuring them all,
and the fact that many are likely to be highly correlated, both temporally and spatially.
The Cebu data provide an unusually rich amount of information regarding a wide range
of community characteristics, in addition to the facility data, to control for these
inﬂuences. These variables were described earlier in the data section and were included
in the speciﬁcations presented in Tables 6.1 and 6.249; results of these effects are
presented in Table C1 of Appendix C.

Unconditional demand for curative health care is reduced by the community being
high in elevation (these are poor), greater barriers to water access, and high present and
lagged infant formula prices. These factors probably operate primarily through negative
short-term demand elasticities. Unconditional demand is also reduced by factors which
relate positively to income and better health environments and are likely to have negative
illness elasticities: the community having paved or concrete roads and a greater
proportion of households with sanitary garbage disposal methods and modern toilet
facilities.

The experiment of removing the community variables was performed to explore
what impact this would have, if any, on the health facility results. If their exclusion
changes the estimated impact of health care characteristics on utilization, this is an

indication that detailed community effects should be accounted for when estimating

 

“9 The results from Table 6.1, the ﬁrst column of Table 6.2, and Table C1 of
Appendix C are from the same regression.

73

determinants of demand for health care services; otherwise, estimates of price and quality
inﬂuences could be biased in unpredictable directions.

The results of this exercise are displayed in Table 6.3. The ﬁrst column of this
table is a similar to the regression in the ﬁrst column of Table 6.2, only the non-health
care community variables have been removed, and so on the the second and third
columns. It is observed that in the distance-only and distance-fee versions, signs and
signiﬁcance of these parameters do not change. Results in the version with quality,
however, are affected in important ways. Most notably, each of the three price effects is
now signiﬁcant, with public and traditional remaining positive and private changing to
negative. The public price parameter increases only slightly, but its z-statistic rises from
1.85 to 3.06.50 The traditional price effect goes from essentially zero to being slightly
positive and signiﬁcant.” The change in the private price parameter is not large in
magnitude, but the sign switches from just greater to just less than zero and is now of

consequence statistically.”

 

5° Correlation matrices reveal that higher public prices are associated with higher
food prices and poorer community infrastructure, such as frequent water shortages,
greater distance to water sources, a lower proportion of households in the community
with running water, sanitary garbage disposal, reﬁigerators, and improved toilet facilities.
Poorer infrastructure could increase demand for services through higher illness levels due
to worse sanitation conditions.

5' Traditional fees, unlike public fees, are correlated with better community
sanitation and infrastructure. Perhaps these community inﬂuences increase the likelihood
of a traditional visit through their income effects, as opposed to illness reducing effects.

52 Private facility fees are associated with better community sanitation and

inﬁ'astructure. It is possible these community aspects may reduce demand by lowering
morbidity.

74

These changes indicate that non-facility community characteristics have important
inﬂuences on other determinants of demand, and when omitted, lead to biased results.
While the direction of the biases could not have been determined a priori, in this example,
exclusion of non-facility community controls results in an upward bias in the public and
traditional fee coefﬁcients, and a downward bias in the effects of private fees.

For the health care quality effects, ORT supplies is the only public feature which
remains important after removing the community controls, and its effect is virtually
unchanged. Public personnel levels and vaccine availability lose their positive effects.
This downward bias could signal that without non-facility community controls, these
quality measures may capture community inﬂuences which decrease the likelihood of a
visit. These two facility aspects are associated with better infrastructure, which could
decrease demand by improving the disease environment in the community.53

The private characteristics which remain inﬂuential are the positive effects of
personnel level and nurse ratio.54 Private doctor ratio loses its signiﬁcant positive effect,
again, indicating that without the community controls, the parameter be biased toward
zero because it captures omitted negative inﬂuences on demand.55 The provision of
family planning and postnatal services in private facilities now have important negative

impacts on demand for child curative services; the change in these coefﬁcients may also

 

53 Public ORT supplies are correlated with poor sanitation.
5“ Both are correlated with better community sanitation and infrastructure.

5’ Private doctor ratio is associated with worse community sanitation which could
raise illness levels.

75

signal that they now capture the effect of omitted community inﬂuences which reduce the
probability of a visit.56

Neither of the traditional quality variables becomes signiﬁcant when non-facility

community effects are removed.

56 These two facility variables are associated with better sanitation and
infrastructure which may reduce demand by lowering morbidity.

76

W

The provider choice model was estimated using the several different speciﬁcations,
which were described in the empirical modeling section. Discussed ﬁrst is the baseline
speciﬁcation: the multinomial logit model of the choice of public care, private care,
traditional care, or no visit. The model has the attractive feature that it allows facility
effects to vary by alternative. We also include a very comprehensive set of non-facility
community variables to control for inﬂuences whose exclusion could bias the facility
results. Alternative speciﬁcations are then presented and discussed.

For most of the following speciﬁcations of the facility choice model, the omitted
category for comparison is "no health care visit"; this implies that the coefﬁcients for
each alternative give the effect of that variable on the chances of visiting a facility of this
type versus not having a health care visit. Similar to the logit results for the "any visit"
model, standard errors are corrected for the intra—cluster correlation which arises from
repeated observations on the mother-child pairs over the twelve survey rounds. This
itself is an unusual extension to the multinomial logit model with panel data; not
adjusting for intra-individual correlation produces incorrect and misleadingly low

standard errors.

INDIVIDUAL AND HOUSEHOLD INFLUENCES
MAIN RES UL T S

Across each of the speciﬁcations, the impacts of individual and household
characteristics are very stable and will therefore be discussed ﬁrst. Table 6.4 shows a

large number of important individual- and household-level inﬂuences on the choice of

77

 

care. Starting with individual child-level variables, male children have a higher
likelihood of being taken for a private provider visit. Private is the most expensive and
highest quality type of care. Given that overall reported illness based 24-hour recall of
symptoms does not vary by child sex for the sample as a whole,57 this result may imply
that male children are treated preferentially for health care investments. This could arise
for several reasons: parents may have the perception that boys are more vulnerable to
illness since their reported morbidity levels are slightly greater in the months
immediately after birth; alternatively, if boys are more likely than girls to contribute to
parent security in old age, perhaps their health needs are attended to ﬁrst.58

Male children are also more likely than girls to have a traditional health care visit.
Traditional services are also relatively high-priced compared to public care and may offer
certain special types of treatments not available in modern centers. Similar explanations
of preference for males or the perception of male child vulnerability to illness could hold

here is with private visits.59

 

57 See Footnote 6 in Section 1 for a description of the differences in reported
morbidity levels for boys and girls by age and by type of symptom.

5“ Alderman and Gertler (1997) also ﬁnd in rural Pakistan that conditional on an
illness being reported, there is a tendency to use high—quality providers more often for
males than females. Given that theirs is an unconditional model, if it holds that bias
exists against reporting girls' illnesses, then only the very sickest girls will appear in their
sample. If this were the case, it is possible that the high-quality pro-male bias could be
even stronger in an unconditional version of their model.

59 In the conditional model of Alderman and Gertler (1997), boys' demand for
traditional services is no greater than girls'. However, in their area study area, traditional
practitioners have the lowest prices among all the health care alternatives, which is not
true in Cebu.

78

Chances of a visit to each kind of care increase signiﬁcantly with child age up
through the sixth month after birth.60 The strongest effect of this ﬁrst linear segment of
the child age variable is for traditional visits, followed by private, then public.

For children older than six months, age does not affect demand for traditional care,
but it reduces demand for public and private care drastically. This is despite the fact that
illness levels stabilize, but do not decline after six months of age. The nature of child
morbidity does change, however, with age: "other" illnesses become more prevalent and
diarrhea and FRI rates level off after the child's ﬁrst six months. It is possible that
traditional services are deemed more efﬁcacious or appropriate for these "other" types of
illness. Regarding modern treatments, perhaps parents become more accustomed to their
child having diarrhea or FRI and may not be as likely to take them for a modern health
care visit relative to when the child was a newborn; it is possible that the child may be
viewed as being more able to withstand the experience of an illness episode after this
perinatal period. Alternatively, parents may have learned to cope with these child
illnesses by treating them at home.

Maternal education has strong positive effects on the probability of a private visit;
the impact of additional years of schooling up to the ﬁfth are particularly large. The ﬁrst
linear piecewise segment of the maternal height variable (up to 150 cm) also has a
noticeable positive inﬂuence on modern private and public visits, indicating that this
variable is probably capturing accumulated human capital and family background

characteristics of the mother not picked by education. Maternal height over 150 cm

 

6° This is the period when reported child illness levels accelerate most rapidly.

79

lowers the probability of a public visit, possibly through an income or illness reducing
inﬂuence of this human capital/family background factor. Years of paternal education
beyond the ﬁfth, (i.e., the second linear piecewise segment of the variable) have negative
impacts on the chances of a public or traditional, but not a private, visit. This may reﬂect
paternal preference for fewer health care visits to lower quality facilities.

Insurance coverage and household assets up through the ﬁrst tercile value greatly
enhance the probability of a private visit. Assets reﬂect higher household income and
perhaps preference for higher quality care. Insurance coverage probably also reﬂects
greater income and preference for quality and may reduce the high price of private care.

Additional infants, particularly males, in the household younger than the index
child decrease the chance of an index child's visit to the two more expensive providers,
private and traditional. This result could arise for a variety of reasons. Additional
children could tighten household income constraints, thus reducing number of visits to
more expensive providers for all children. However, because the effect is greater for
additional male than for female infants, this is probably not just a simple resource
crowding story, but may reﬂect a preference for newborn males. Alternatively, reported
illness levels increase sharply during the ﬁrst few months of life for all infants, and boys'
reported rates rise slightly faster than girls' during this period. Parents may therefore be
acting rationally by investing ﬁrst in the health care of the youngest male children
because they may be (or at least may be perceived to be) the most vulnerable to illness.

Demand for private services increases with additional adult females in residence;
this may reﬂect a preference on the part of adult females for higher-quality child health

care services.

80

Additional elderly males in the household reduce child visits to the two more
expensive types of care, private and traditional. Their presence may impose time and
economic burdens on the household and therefore reducing allocations to child health

investments.

ROBUST NESS CHECKS ON HUMAN CAPITAL AND ASSET RESULTS

Robustness checks were performed on the individual and household effects. This
was accomplished by estimating the facility choice model with community-time
interaction dummies used in place of the facility and community variables to capture all
community- and time-varying inﬂuences on demand. We ﬁrst included all individual and
household level variables. We then dropped the health insurance coverage dummy to see
whether this would alter the asset and human capital results. Then the household
composition effects were removed to examine whether this would change the inﬂuences
of human capital, assets, or insurance. Finally, both insurance and household
composition were dropped to examine the robustness of the impacts of human capital and
assets.

The results, presented in Table 6.5, reveal that the exclusion of health insurance
coverage results in no changes in individual and household impacts on the demand for
public care; for private services, the positive effect of additional paternal education over
ﬁve years becomes signiﬁcant at the 10 percent level, although the magnitude of the
effect is still not large. No changes are observed in the determinants of demand for

traditional visits.

81

With household composition dropped, human capital, asset, and insurance results
for public care are not altered. The positive inﬂuence of the second asset spline (values
above the ﬁrst asset tercile) on the probability of using the two more costly types care,
private and traditional, is strengthened.

When both insurance and household composition are dropped, the negative impact
of the second paternal education spline on a public demand is strengthened slightly, as is
its positive impact on the probability of a private visit. The positive effect of the second
asset spline on private and traditional care becomes slightly more statistically signiﬁcant.

These results indicate that the health insurance and household composition
coefficients reﬂect to some degree the impacts of human capital and household assets, but
including them does not greatly alter the inﬂuences of these household human and

economic resources.

HEALTH FACILITY INFLUENCES
BASELINE MODEL

We did the same experiment of including the facility attributes in successive steps
here as we did in the binary logit "any visit" model. Results for the distance-only model,
presented in Table 6.6, indicate very strong negative own—effects. Each of the facility 2-
statistics is greater than 4.2; the magnitude of the effect is largest for traditional visits,

followed by public, and then private.“

 

6' Alderman and Gertler (1997) also ﬁnd that health care providers which are
considered inferior sources of care have the highest price elasticities.

82

With user fees added to the equation, distance does not lose any inﬂuence. All user
fees are near zero and insigniﬁcant; the public user fee coefﬁcient, although not
signiﬁcant, is positive and slightly larger in magnitude than the other two fee effects.

When facility quality variables are added, all the distance results remain negative
and very signiﬁcant. The public user fee effect is still insigniﬁcant and its coefﬁcient
moves closer to zero, possible indicating that some of its previous positive inﬂuence may
have been capturing facility quality inﬂuences. It may continue to have a positive sign
because of service aspects not included in the analysis, such as those relating to the
process of delivering care: factors such as facility cleanliness, staff motivation and
attitudes, etc.

The price effect for traditional care becomes positive and signiﬁcant when provider
attributes are included; this result was unexpected. Very little information pertaining to
the quality of traditional services was collected, so the positive effect may arise from
traditional provider attributes which are still omitted from the model, or from correlation
between traditional quality and other regressors.

Among the quality inﬂuences, higher numbers of personnel and ratio of doctors in
public facilities raise demand for public services. Private personnel do not inﬂuence
demand at private establishments. The personnel measures relevant to traditional
practitioners, the education level and the formal training indicator, have positive effects,
particularly that of training.

Drug supplies show strong inﬂuences on public, and to a lesser extent, private
demand. Currently available ORT treatments and vaccines at public facilities raise

demand for public care, while public availability of intravenous diarrhea treatments

83

reduces it. The ﬁrst two effects are what we might expect; however, the latter result is
not. The fact, however, that the drug variables are deﬁned as those currently in stock
could be driving this result. The availability of a drug at any particular time reﬂects the
interaction of both supply and demand factors; if a certain drug is out of supply it may be
due to excess demand for it. For instance, if public clinics with intravenous drugs are in
areas with high demand--poorer areas with high diarrhea levels--their stock may be
diminished quickly; this would produce a negative correlation between demand for
services and availability of drugs (Mwabu, et al., 1993). The signiﬁcant negative effect
of intravenous diarrhea treatments may, therefore, indicate that they are high demand at
public health facilities in high-diarrhea areas.

The supply of intravenous diarrhea treatments at private facilities increases the
chances of a private visit. For the other four indices representing current availability of
expendable supplies at private centers, each had important negative effects on demand;
these were availability of simple ORT treatments, child vaccines, and family planning
methods. It is possible that these items are in high demand in private clinics; the
depletion of such supplies could be due to high rates of utilization of these services at
such facilities.

The full set of results for the baseline ﬂexible facility choice model, including
parameters for all the community infrastructure and sanitation variables are presented in
Table C2 of Appendix C. The table reveals that unconditional demand for public care is
greater in communities which are islands, which have more frequent water shortages; it is

dampened by high elevation status, the presence of improved roads, a bank, longer time

84

to the nearest water source, higher lagged infant formula prices, and a higher proportion
of households with sanitary garbage disposal.

Demand for private care is increased by lagged community vegetable oil price; it is
reduced by greater distance to nearest water source, the prevalence of modern toilets, and
by lagged rainfall levels.

The likelihood of a traditional visit is increased by community island status, the
presence of a bank, a higher proportion of refrigerator owners, and higher lagged corn
prices. Traditional demand is decreased, on the other hand, by more frequent water
commrmity shortages, greater time to closest water source, and higher current cooking oil
prices, and a higher average number of households with sanitary garbage disposal

methods and modern toilets.

EFFECTS OF REMOVING NON-FACILIT Y COMMUNITY CONTROLS

As in the "any visit" model above, we applied the experiment of removing non-
facility community inﬂuences to investigate what effects, if any, this would have on the
health care quality ﬁndings.62 Table 6.7.1 presents the set of regressions where provider
characteristics are included in successive steps, with the community controls excluded
each time.

In the distance-only regression, the distance effects become more negative and
signiﬁcant for public and private visits relative to when non-facility community

inﬂuences are controlled for; the downward biases in these effects imply that the modern

 

‘52 Results on individual and household impacts are virtually unchanged, so they are
not discussed.

85

distance results may now capture omitted non-health-facility characteristics of the
community which have important negative impacts on the probability of a modern visit.

The impact of distance on traditional visits, however, is biased in a positive
direction: its former negative parameter is reduced both in magnitude and signiﬁcance to
virtually zero. Traditional distance could now be capturing omitted community
inﬂuences which as a group have a positive impact on demand for traditional services.63

In the distance-user fee version, distance has similar results to those we observed
with the community controls included. Private and traditional user fees are also very
close to what they were with the controls. The public user fee result, however, changes
dramatically: its parameter is now negative, large in magnitude, and signiﬁcant at the
10% level.

This ﬁnding of a large downward bias in the public user fee effect when other
community inﬂuences are dropped from the analysis has important implications for how
public price effects from these types of studies are interpreted. Without the community
controls, we would conclude that raising public fees would have a very large negative
impact on demand for public health care services. However, when infrastructure and

other characteristics of the community, which may capture income levels and aspects

 

'53 Distance to the nearest of each facility type is associated negatively with
community infrastructure and sanitation. Poorer infrastructure is likely to be related to
low incomes which would have the effect of reducing demand for services; on the other
hand, lack of infrastructure is likely to be related to poorer hygienic conditions which
could increase demand for health care by raising morbidity.

86

related to the disease environment, are taken into account, we see that raising fees on the
margin is not likely to inﬂuence overall utilization“ of public services.65

In the model with health care quality included, distance is still a major inhibitor of
public and private visits, as in the baseline model. Similar to the baseline results, private
fees are close to zero. Now, instead of being positive and signiﬁcant as in the baseline
model, traditional fees here are close to zero in magnitude and signiﬁcance.

Similar to the previously described distance-fee speciﬁcation, the public user fee
result is sizeable, negative and signiﬁcant at the 10% level, unlike the ﬁndings from the
baseline model where its magnitude was small and positive and its impact not statistically
signiﬁcant.“

As discussed above, this non-robust price effect could have important implications
for the interpretation of results from other health care demand studies. Very few analyses
of this kind have the detailed controls for non-facility community inﬂuences that we do

here. At best, district- or region-level dummies are employed to account for any effects

 

6" The effects on utilization among different income and demographic groups is
analyzed in a later section.

65 As presented in Table 6.7.2, even when detailed community characteristics are
replaced by dummies for the eight municipalities (the administrative unit just higher than
the community level) which comprise the study area, the public user fee effect is
negative, large in magnitude, and is statistically signiﬁcant at the zero percent level.

66 For this version of the model, replacing the community characteristics with the
eight municipality dummies, results in a public user fee effect which is negative, large in
magnitude, and is statistically signiﬁcant at the ﬁfteen percent level.

87

beyond the individual or household level. Therefore price responses from other health
care demand studies could also potentially be affected by this issue.67

None of the quality results changed sign, although a few changed in magnitude and
signiﬁcance. Vaccine availability at public clinics loses importance, while public family
planning now appears to signiﬁcantly increase demand. For private visits, all the drug
indices become unimportant. The negative inﬂuences of doctor and nurse ratio become
signiﬁcant.

In conclusion, because health care attributes are likely to be highly correlated with
other community characteristics, the interpretation of price and quality effects on demand
from models which do not include non-health—facility community controls, or which
employ only district-level dummies to account for community inﬂuences, should be

interpreted with caution.

CONDITIONAL LOGIT SPECIFICA TION
In addition to the issue of other potential non-health facility community inﬂuences,
many studies use a conditional logit approach in which each health care characteristic is

forced to have the same impact on demand for each type of provider. Given, however,

 

(’7 One caveat, however, is that much of the research on this topic has used distance
as the only price variable in the model. In this study, although the negative effects of
distance on public and private care demand are substantially weakened when other non-
health-facility community inﬂuences are controlled for, they remain statistically
signiﬁcant. Therefore, even though studies that do not measure price using both distance
and user fees are less complete, if the distance inﬂuences are large to begin with, it is
possible they could remain important even after controlling for non-facility community
inﬂuences. This, however, is an empirical question which can only be answered by
including both distance and user fees in the model, with the non-facility community
inﬂuences ﬁrst kept and then removed.

88

the wide variation in the nature of the service types, e.g., personnel levels and training,
drug availability, and inevitably other unmeasured aspects of service, one can make a
strong argument that care from the different segments of the health care market can
reasonably be considered to be different goods. For example, an increase in the doctor
ratio in an isolated clinic is likely to affect the nature of that service in a way which may
impact demand very differently from the same percentage increase in doctor ratio at a
modern urban hospital. Therefore, in the baseline model used for this study attributes
which are relevant to each type of service (distance, price, number of personnel) are
allowed to inﬂuence demand for each service type differently; furthermore, different sets
of attributes are allowed to impact the probability of a visit at each.68

Using a conditional logit model similar to others estimated in this literature, we
demonstrate how dramatically our provider attribute effects these diverse services are (a)
forced to have the same set of elements describing them, and (b) each element is forced to
have an identical impact on the demand for each type of care.

Individual and household effects are virtually unchanged from the baseline model
so are not presented here. Table 6.8 shows the results of imposing identical effects of the

health care attributes on each type of service, and of excluding non-health-facility

 

6" Analytically, this feature of the conditional logit model implies that each element
of the choice set must contain the same set of explanatory variables; therefore, in
transforming the facility data to be used for the conditional logit model, either (a)
descriptors that do not apply to a particular choice must be included for that choice with
the value set to zero, or (b) only the facility variables common to all choices are retained
in the analysis; for this study (b) would imply that only the price and distance variables
are kept.

89

community controls.“’9 The previously observed differential inﬂuences of quality and
price on demand for different types services are masked by this type of model; changes
from the baseline results are quite large.

Relative to the baseline ﬂexible MNL model, user fee is strong and negative for
each choice; previously each own price effect was either zero (the case of modern
facilities), or positive (the case for traditional care). Doctor and nurse ratios are positive
and inﬂuential for all types of care, whereas in the baseline version only public doctor
ratio increased the chances of a public visit. The vaccine availability effect is now
positive across the board; in the baseline model only vaccines at public facilities
increased demand for public care. Family planning offerings are negative and signiﬁcant
for all types of visits, as opposed to the baseline results where only those at private
facilities reduced the probability of a private visit. Access to intravenous diarrhea
treatments has no effect on visits now. Training of traditional providers increases the
chances of a visit to each facility type, instead of just to traditional facilities.70

To conclude, results from our baseline ﬂexible MN L model indicated that when
differential inﬂuences of price and quality on demand are allowed to vary by the type of
care, there are indeed differential demand responses by service type. Furthermore, the

ﬂexible speciﬁcation allows for the possibility of having a different set of quality aspects

 

69 It should also be noted that standard errors are NOT corrected for intra-individual
correlation due to repeated observations on individuals. Therefore, the z-statistics
presented are higher than they should be.

7° If we compare these results instead to the ﬂexible MNL model where community
controls are also excluded (on the basis that they are more comparable with this model),
we again ﬁnd that many facility effects which varied by provider type in the ﬂexible
model, change sign and signiﬁcance using the conditional logit speciﬁcation.

90

affect the demand for each kind of care. Therefore, forcing different ﬂavors of health
care to be inﬂuenced by the same set of attributes, and imposing the restriction that each
of these attributes has the same effect on every kind of service, is unrealistic at best and
could be highly misleading. Furthermore, it puts unnecessary limits on the insights
which can be drawn concerning how prices and quality affect demand for care from

different segments of the health care market.

NESTED M ULT INOMIAL LOGIT SPECIFICA T ION

Estimated next is the ﬂexible two-level nested multinomial logit model with no-
cross effects, as described by equations (15)-(17) in Econometric Methods section. The
results are presented in Table C3 of Appendix C. Turning ﬁrst to the inclusive value, its
parameter is 0.36. The result that the inclusive value parameter is between zero and one
tells us that the data reject neither the distributional assumptions nor the functional form
of the utility function; therefore, the model is consistent with utility maximization
(McFadden, 1981). The fact that it is less than one implies that the error terms in the
conditional utility functions for the provider alternatives are correlated, implying that the
simple MNL model is rejected in favor of this speciﬁcation.7l The magnitude and
signiﬁcance of this parameter is also very close to the results found for children in rural

C6te d'Ivoire and rural Peru by Gertler and van‘der Gaag (1990).

 

7' We also experimented with the three-level nested MNL model that was laid out
in the empirical modelling section: visit or not--modern or traditonal--public or private.
We found that when public and private care were grouped into a "modern care" branch
separate from traditional care, we could not reject that the inclusive value parameter on
this branch was equal to one. Therefore, the three-level tree was collapsed into the two-
level tree estimated here.

91

Because the choices are now divided into two levels, (1) a visit or no visit, and (2)
the type of care chosen conditional on having a visit, there is no common base category
for each nest of the decision. In the ﬁrst level, the coefﬁcients measure the change
relative to no health care visit; in the second level, they measure the change relative to
having a traditional visit. Because the base categories differ, the estimated coefﬁcients
from the nested model cannot be directly compared to those from baseline MN L model.
For this reason, marginal effects for the unconditional probabilities of a visit to each
facility type were calculated for both the MNL and NMNL models using equations (16)-
(1 7). These are presented in Table 6.9.72

For public demand, we see that the marginal effect distance is reduced slightly in
the nested version. Public user fee remains positive and increases in magnitude. (Its
NMNL coefﬁcient, however, is still not statistically signiﬁcant.) The public quality
results are not altered to a great extent. Nurse ratio is the only effect which changes
direction, from positive to negative, and its NMNL effect is relatively large. The
negative impact of availability of intravenous diarrhea treatments is increased by sixty

percent.

 

72 In the MNL model, the base category for comparison was not a health facility,
and hence did not have measured health care attributes which uniquely described it.
Therefore, there were no characteristics of the base category which affected the indirect
utility functions of the other alternatives.

In the NMNL model, however, the base category for this subset of alternatives is
traditional care. The characteristics of this facility type to enter the lower level of the tree
where facility choices are made conditional on there being a health care visit. Because
this omitted base category is a health facility, which has measured attributes unique to it
and which are not found in the indirect utility functions for the other alternatives, its
characteristics enter the estimated equations for the other choices. (See Appendix B for
more details.) Because parameter estimates are obtained for the attributes of this base
facility, their unconditional marginal effects can be estimated.

92

The inﬂuence of private distance on private demand is very close to zero in the
nested speciﬁcation, instead of being negative as in the baseline model. The sign of the
private user fee effect changes from negative to positive, but remains essentially at zero.
None of the private quality effects changes direction; however, the magnitude of each,
with the exception of the postnatal service indicator, is reduced. The negative impacts of
vaccine and family planning availability are most affected: both marginal effects are
smaller.

Nesting alters the impacts of some traditional provider characteristics. The impact
of distance does not change. However, the positive inﬂuence of user fee is reduced to
one-third its original value. The impact on demand of educational level of traditional
providers in the community does not change with nesting, but the positive inﬂuence of
their having participated in any formal health training is increased by sixty-ﬁve percent.

Some of the strong individual and household ﬁndings in the MNL model are
different with the nested structure.73 Maternal and paternal non-residence for the ﬁrst two
years of the infant's life appear to increase more the chances of a public or private visit.
For private visits, the positive inﬂuences of being male, having health insurance, higher

household asset levels, and adult females in residence are each diminished. The negative

 

73 Note that even though individual and household-level variables are included in
both levels of the nest (i.e., they inﬂuence whether there is a health care visit, and choice
of facility conditional on there being a visit), their unconditional marginal effects are not
given for the traditional versus no visit choice. This is because traditional facility is the
normalized base category in the conditional choice nest, which means that coefﬁcients for
individual and household variables are not available from this nest for which to calculate
the ﬁnal unconditional marginal effects for the nested speciﬁcation.

93

effects of younger children and older males in the household are also reduced. The child
1-6 month age spline changes from positive to negative.

These results are based on a model which may be deemed more realistic
analytically because it allows correlation in the error terms of the three health facility
choices, thus allowing the choice among the three provider types to be more similar to
each other than to the choice of no health care visit. However, self-selection issues arise
with this speciﬁcation because the results for the three-choice health facility level of the
tree are based only on the subsample of children in each round whose mothers took them
for a health care visit during the two-month interval preceding each survey. This source
of selection may result in biased estimates for the sequential nested multinomial logit
model. Direct comparison of the results to the MNL model where selection bias is not an

issue should therefore be done with this caveat in mind.

FULL CROSS-BF F EC TS SPECIFICA T ION

The ﬁnal variation of the model is the "ﬁrll cross—effects" MNL version in which all
facility variables are allowed to enter each indirect utility ﬁmction. Results are displayed
in Table C4 of Appendix C. Individual and household effects are virtually unchanged
from the baseline so are not presented. Several own-facility impacts change when cross-
inﬂuences are allowed; the baseline results for public facilities are more robust than those
for private facilities. Among the public attributes, the positive effects of price and family
planning increase in magnitude and become signiﬁcant; number of personnel is no longer

signiﬁcant.

94

Among the private impacts, price and provision of postnatal services both change
direction (from negative and from positive, respectively) but neither is signiﬁcant either
here or in the baseline model. Doctor ratio switches from negative to positive and
becomes signiﬁcant; number of personnel gains signiﬁcance but is still close to zero in
magnitude. Family planning index loses signiﬁcance and changes from negative to
positive.

The positive effect of formal training of traditional providers loses its signiﬁcant
impact.

The cross-effects are in some cases of the expected sign and in other cases not."
For instance, increased public user fees now cause the demand for traditional services to
rise, but they decrease the demand for private care. Greater numbers of public facility
personnel lower the chances of using the other two types of care as we might expect;
however, public availability of intravenous diarrhea treatments and vaccines is now
related to increased demand for private and traditional services.75

Private provision of postnatal care decreased demand for public services, but raised

the probability of a traditional visit. Private personnel and nurse ratios also had

 

7“ In another speciﬁcation, interesting cross-provider effects were seen in a
distance/user fee regression before quality was included. Distance to a public facility
increased the probability of using a traditional provider; analogously, distance to
traditional raised the likelihood of a public visit. Public user fee also had a strong
positive impact on traditional visits. These effects diminished somewhat when quality
was added.

75 As discussed previously, higher current supply of drugs at public facilities may
be an indication that services for which these drugs are applicable may not be in very
high demand at public facilities. If so, this result is not as counterintuitive as it ﬁrst
appears to be.

95

unexpected positive effects on traditional visits. Education of traditional practitioners
decreased chances of a private visit, but raised the likelihood of a public visit.

The unexpected cross-effects may arise from high correlations in certain
characteristics between facility types. For instance, public and private user fee are highly
correlated, as are distance and many of the personnel measures. In our results, each of the
signiﬁcant cross-distance effects are negative. Ex ante, we may have expected greater
distance to say a public provider to have increased demand for private or traditional
services; this, however, did not turn out to be the case. Dor and van der Gaag (1987)
obtain similar results when comparing models similar to this one which allow cross-
effects and then restrict them to zero; their unexpected results are attributed to the actual

and spurious correlations in their health facility data.

POLICY SIMULATIONS

Using the results from the baseline ﬂexible MN L model (where facility effects are
permitted to vary by provider type, but with cross-effects restricted to zero), simulations
were performed to analyze the effects of hypothetical changes in public health care
characteristics and in household attributes on the demand for health care services. Much
of the policy discussion in this area focuses on the likely effects of reforms in the public
health sector on utilization, both overall and among poor households. Therefore, several
of the simulations involve changes to possible public policy levers; furthermore, separate
sets of predicted responses are derived for the poorest households, for those who are

better-off , and for the sample as a whole.

96

Simulated probabilities of utilization are obtained by ﬁrst calculating the baseline
probabilities using the estimated parameters derived the baseline regression model and
the actual value of each explanatory variable; these probabilities are averaged. Next, the
impact of a change to public facility or household attributes are examined by changing
the value of the explanatory variable of interest, recalculating and averaging the
probabilities, and comparing these results to the baseline probabilities.

It should be noted that the simulated responses to these hypothetical changes are
derived using parameters obtained from the original regression model; therefore, the
approach is valid only for marginal changes in the factors that inﬂuence health care
visits.

Table 6.10 presents the results of these simulated demand changes for (a) all
households, (b) households in the lowest asset tercile, and ( c) households in the upper
two asset terciles. Reading across a row gives the probability of choosing each provider
for a given asset class; the probabilities for each row sum to 100 within each asset class.

We observe that making ORT available at each sampled public facility increases
overall public utilization by 17 percent. Low-asset households respond less to this
change than high-asset households, the estimated demand increases being 7 percent and
24 percent, respectively. The to public care among both high- and low-asset households
comes from approximately equal percentage decreases in private, traditional, and no
health care use.

Completely eliminating public ORT supplies reduces overall predicted public
utilization by one-half. As with the ORT improvements, low-asset households respond

less to this change. Their probability of a public visit is reduced by about one-third,

97

compared to a 54 percent reduction for those of higher SES status. For both asset classes,
this policy change would result in equal percentage increases in private and no health
care, and a slightly lower percentage increase in the use of traditional care.

Increasing the available range of child vaccines at public centers raises the
predicted probability of a public visit by 29 percent. Disparities in responses by asset
class are not as strong as they were for ORT changes; the increase for poorer households
is 25 percent and for richer households is 31 percent. Percentage reductions come equally
in private, traditional, and no care use, and the reductions were similar regardless of asset
category.

The effect of improving the number of family planning methods available at public
facilities is to raise the overall probability of a public visit by 18 percent. Reactions by
asset group differed however; this caused poor household to decrease their rate of public
visits by 11 percent, while it raised the probability of a public visit by a better-off
household by 36 percent. For the poor, percentage increases in private, traditional, and
no care in response to this policy change were approximately equal. For high-asset
households, the probability of using private, traditional, or no care we each reduced by
about the same percentage.

Simultaneously enacting the three drug supply improvements increases the
probability of public utilization by 74 percent overall, by 19 percent for the poor, and by a
full 106 percent for better-off households. For the overall sample and for the upper-
income group, the percentage increase for the simultaneous improvements is greater than
the sum of the increases due to each of the three individual improvements; it is

approximately equal to the sum for the poor. In the upper-income group, the new users

98

are drawn largely from those previously having a private visit, a slightly lower percentage
from those who had no visit, and a still lower percentage from former traditional users.
There were approximately equal percentage reductions in private, traditional and none
care use among the poor.

For the 25 percent of households furthest away from public facilities, we
simulated reducing their barriers to access. In terms of policy actions, this could
correspond to (a) building facilities only in under-served areas or (b) improving
transportation infrastructure to reduce travel times. For the twenty-ﬁve percent of
households furthest from a public facility, reducing their distance to that of seventy-ﬁfth
percentile results in a ﬁve percent increase over the baseline probability of using public
care, with the increase coming from former traditional and non-users. While the response
among low-asset households is less, at two percent, the increase comes almost solely
from those who did not use care before, thus increasing this groups overall use of health
care. For higher-asset households an eight percent increase is seen, with new users being
drawn at equal percentages from private, traditional, and no care.

Raising mother education by one sample standard deviation (3.7 years) is
expected to raise public demand by approximately 4 percent for the pooled sample. The
increase among the poor, however, is much greater: their predicted probability of a
public visit is increased by 13 percent, and the great majority of these new users are
drawn from the group who had no health care visit before. Among upper-income
households, the increase is very small at one percent.

Increasing the value of household assets by one sample standard deviation shifts

overall utilization by reducing the chances of a public, and to a lesser extent, a private

99

visit; it increases the likelihood of using private services and decreases only slightly the
probability of not using health care. For the poor, increasing the value of assets boosts
them into the non-poor category. Because of the very large positive effect of the ﬁrst
asset spline on private demand in the baseline model, the result of this simulation is that
all the poor switch from public, traditional, and no care, into using private services.
Upper-income households would also use more private care, the percentage increase
being 4 percent.

In sum, the policy simulations reveal that improving aspects of quality at public
facilities which can directly beneﬁt children, speciﬁcally ORT availability and vaccines,
raises demand for public health care, especially among the poor. The new users are
drawn from former private, traditional and non-care users. Increasing the range of
available family planning methods at public facilities, a policy which in theory could
increase the likelihood of a child visit if the adult taking the child (usually the mother)
highly values concurrent access to these services, raised public demand only among
higher-asset households. The shift came as a result of reductions in private, traditional,
and no care. Simultaneously expanding all three of these supplies had huge positive
impacts on the upper asset group's public utilization, and positive but less striking effects
for the poorest; percentage reductions were seen in each of the other three choices for all
asset classes.

Reducing distance to public facilities led to positive but small increases in public
utilization, and the effect was smaller for the poor. Increasing maternal education had
large inﬂuences on public sector utilization among the poor; most new users were drawn

from the pool of former non-users of health services. Raising the level of household

100

aunt
especia

SCCIOI.

assets resulting in enormous increases in the demand for private health care services,
especially by the poor, mostly as a result of former non-users being drawn into the health

sector.

101

Section 7

RESULTS FOR THE DEMOGRAPHIC AND SOCIOECONOMIC SUBGROUPS

In setting health sector priorities, it is important to know if prices and quality of
care affect demand differently depending on the speciﬁc demographic or socioeconomic
group in question. Often public services are offered and/or subsidized with the intention
of beneﬁtting a particular type of individual. The aim may be to reach those with the
lowest degree of access to the service, or those who could be expected to beneﬁt most
from using the service. In this study, relevant groups who may be targeted for the
provision of basic curative health services include the poor who are generally under
served; low-educated mothers who may have low utilization of services because, besides
being poor, may not be aware of the potential health beneﬁts of use; female children who
may be discriminated against for health investments; or children in a particular age group
who may have special health care needs due to their particular stage of physical
development.

An important element in designing a strategy to target any particular group of
individuals is knowing the determinants of service demand among that group. It is
possible, and in some cases expected, that different types of individuals will have very
divergent responses to price and quality changes. If policies are formulated based on
responses of the average consumer, certain types of potential beneﬁciaries, some of
whom may in fact be the primary targets of the intervention, may be affected differently
than health care planners expected when they designed the intervention. For this purpose,

we estimate demand responses for subsarnples of individuals, disaggregated by four

102

exogenously determined characteristics of interest: (1) sex of the index child, (2)
education level of the index child's mother, (3) value of household assets, and (4) age of
the index child. This provides us the opportunity to investigate whether demand for child
curative health care investments are more price, quality, and/or income responsive among
certain groups.76

For each subgroup the baseline ﬂexible MNL was estimated. Then, as in the
pooled model, the NMNL speciﬁcation was used to explore the effects of relaxing the
assumption of zero correlation in the error terms across facility alternatives. Community
non-health-facility control variables were then removed to investigate how their exclusion
would inﬂuence the health facility impacts. For each of the demographic cuts, the
changes which resulted from using the nested speciﬁcation and from the removal of the
community variables, relative to those of the baseline ﬂexible model for each group, were
very similar to what was observed for the main pooled model. Therefore, in the
discussion that follows we do not include the details of the ﬁndings from these alternative

speciﬁcations.

MALE VS. FEMALE CHILDREN

 

76 When exploring the possibility of differential demand responses to changes in
public health care and household characteristics, it is important that the sample NOT be
split according a variable that is (a) the outcome of interest, or (b) is jointly determined
with the outcome of interest. For example, stratifying the sample by those with low rates
of utilization versus those with high rates would result in two artiﬁcially truncated
subsamples, each of which would produce biased estimates.

103

The results from the baseline pooled model indicated that male children appear to
be favored for health care investments. The likelihood ratio test of the hypothesis of no
coefﬁcient differences between the male and female subsarnples was not signiﬁcant.77
We will still stratify the analysis by child sex, however, because it will allow us to
explore whether demand for daughters' curative health care is more responsive than sons'
to health care price, quality, or household inﬂuences.

Other health care demand studies for children in the Philippines have not found
promale bias in health seeking behaviors.78 Bouis et al. (1997) ﬁnd that gender did not
affect the probability of seeking any of the alternatives to traditional health care providers
for adolescents with a reported illness in Bukidnon. Using country-wide data, Chin
(1995) ﬁnds among sick children age ﬁfteen or less that boys are less likely to be taken
for formal health care services. In Bicol (Akin et al., 1985) child sex is not found to be an
important determinant of health facility choice for children age 0-5.

Gender differentials in health care investments have been found in other countries,
however. Alderman and Gertler (1997) estimate a conditional model of child health care
utilization for rural Pakistan. They ﬁnd that girls' demand for care from government
clinics and high-quality private clinics is more price elastic than boys'; these effects

diminished, however, with income. In rural Cote d'Ivoire, male sex was found to increase

 

77 Chi-square=136, d.f. 203, chiprob 0.995.

78 For other important non-health care inputs into child health, gender differentials
have been found for the Philippines (discussed in Haddad et al. 1996). Evenson et a1.
(1980) and Chula et a1. (1980) ﬁnd promale child bias in intrahousehold food
distribution; however, Aligaen and Florencio (1980), Haddad, et al. (1993), and Bouis
and Pena (1996) do not. (Haddad, et a1 (1993) adjust energy requirements for body
weight; however, differences in body weight could themselves result from accumulated
past patterns of intrahousehold discrimination in nutrient allocations.)

104

the likelihood of a visit to higher quality doctors for adults, but not for children (Dor and
van der Gaag, 1987).79 Gender was not an important determinant of health care choice
for children in Bolivia (Ii, 1996), in rural Peru or rural Cote d'Ivoire (Gertler and van der
Gaag, 1990), or in Indonesia (Gertler et a1. 1995; Deolalikar, 1993)“.

Each of the demand studies mentioned above (except that of Deolalikar) uses a
sample which is conditioned on a current child illness being reported. If there exists bias
in reporting of illness by child gender, say for instance against females, then only the very
sickest females will even appear in a conditional sample. This would lead to biased
estimates of the determinants of health care demand. For example, even if no sex
differences were found in the conditional model, the fact that girls who may be just as
sick are less likely to even appear in the sample of ill children implies that promale bias
in seeking health care could still exist. Estimates which are not conditioned on illness
status will not be tainted by this form of sample selection which could arise from the
under reporting of illnesses for children of a particular gender.

The unconditional determinants of facility choice for curative health care for male
and female children in our Cebu sample are presented in Table 7.1. Beginning with the
facility-level determinants, we see that the negative effect of distance to a public provider

on public demand is ﬁfty percent larger in magnitude and more signiﬁcant for girls than

 

79 Using a conditional model, Dow (1995a) ﬁnds that adult demand for hospital
care is higher for males, but in an unconditional speciﬁcation male sex was found not
have an impact on demand for hospital care, but to signiﬁcantly reduce the demand for

clinic care. In Kenya, facility choice for adult curative care was not affected by gender
(Mwabu, et al., 1993).

8° Deolalikar (1993) ﬁnds no effect of child gender in either his conditional or his
unconditional demand results.

105

boys, indicating that girls' demand for public curative care is more price responsive than
boys'.81 Private distance, on the other hand, has a signiﬁcant, although very small,
inﬂuence only on sons' private demand.82 Since sons appear to be preferred for use of
private services to begin with, perhaps this causes male child demand to be more
sensitive to price. Traditional demand was negatively affected by distance for both sexes;
demand for daughters' care appears to be much more sensitive, however.

Responses to user fees for public and private health care were not signiﬁcant for
either sex. Daughters utilization of traditional services, however, was positively
associated with traditional user fee. Perhaps this result arises from unmeasured quality
aspects positively correlated with price which could be more important for girls than
boys.83

The signs and signiﬁcance of the quality characteristics of public facilities do not
differ between boys and girls. The impact, however, of private quality on private demand

differs more by sex.“ Private provision of postnatal services has an important positive

 

8‘ Results from the conditional model of Alderman and Gertler (1997) for rural
Pakistan also show that girls' demand for care from government clinics is more price
elastic than boys', although the effect diminishes with income.

82 Contrary to these ﬁndings, Alderman and Gertler's (1997) results, which
condition on morbidity, indicate that demand for girls' high-quality health investments is
more price elastic than boys'. They also ﬁnd that the ratio of female to male elasticities is
greater for high-quality care than for other types of care.

83 Alderman and Gertler also ﬁnd that demand for traditional care for female
children is more price elastic and negative than for male children.

8“ To the knowledge of the author, this is the only study which includes detailed

quality measures in examining the demand for health care investments for boys and girls
separately.

106

impact on boys' visits only. Sons' demand is negatively affected by higher private doctor
ratios. The chances of a private visit for a girl are enhanced by more personnel and the
availability of intravenous diarrhea treatments at the facility, but are reduced by current
ORT stocks, vaccines, and number of family planning methods. As discussed in the
results from the pooled analysis, interpretation of the negative impacts of drug
availability is not straightforward because current availability at any particular time
reﬂects the interaction of both supply and demand factors. If supply of a particular drug
is low it may be due to excess demand for it. For example, if private clinics with ORT
supplies are in areas with high demand, their stocks may be diminished quickly, this
would produce a negative correlation between demand for services and availability of
drugs (Mwabu, et al., 1993). The signiﬁcant negative effect of ORT and vaccines may
indicate that they are in high demand for girls at private health facilities.

For traditional services, years of provider education had no impact. Formal
training had large positive effects on demand for both sexes, with the magnitude being
greater for girls.

Among the individual and household characteristics, the ﬁrst linear segment of
child age (months 0-6) has important positive effects on private and traditional visits for
both sexes, but on public visits only for girls. The second linear segment (months greater
than six) reduces the probability of a public or private visit approximately equally for
both sexes.

The impact of the ﬁrst linear section of maternal education (0-5 years) has no
noticeable inﬂuence on demand for public care; its effect is positive and signiﬁcant,

however, for male, but not female, private demand. It also increases the probability of a

107

traditional visit only among daughters. The second maternal education spline (years over
ﬁve) has no effect on public or traditional care; however it raises the likelihood of a
private visit for sons and daughters approximately equally, but the effect is slightly more
statistically signiﬁcant for girls than boys. It is possible that higher education of mothers
has a more signiﬁcant effect on the demand for daughters' high quality services than for
sons'.85

Maternal age has a signiﬁcant impact on male public visits only and reduces their
likelihood. The ﬁrst linear portion of the maternal height variable (up to 150 cm)
increases the likelihood of private visits for both sexes, but not public visits as it did in
the pooled model. The second height spline decreases demand for public services among
girls only. Index child's mother being the senior woman in the household signiﬁcantly
increases private visits for sons, but reduces the likelihood of a daughter's traditional visit
(perhaps the daughter effect is through a reduction in illness since this variable is negative
for demand each type of care for girls).

The absence of the index child's father during every round of the survey reduces the
probability of male, but not female, traditional visits. The second piecewise segment of
paternal education (years over ﬁve) has a negative effect on the likelihood of public and

traditional visits for sons, but does not affect any type of female care. It is conceivable

 

that more highly educated fathers have a distaste for taking their sons to low-quality care.

 

85 Thomas (1994) ﬁnds a greater impact of parents' education on the health of
children of the same gender in the United States, Ghana, and Brazil. King and Lillard
(1987) and King and Bellew (198 8) ﬁnd that mother's education has a stronger positive
inﬂuence on daughter's education than on son's in the Philippines, Peru and Malaysia.
(Quoted in Quisumbing, 1994).

108

The ﬁrst piecewise segment of household assets (values in the ﬁrst tercile) increase
private demands for both sexes. The second asset spline (values in the upper two terciles
of the distribution) has the effect of reducing daughters' demand for public and traditional
visits. Perhaps richer households prefer their daughters not visit lower-quality health
facilities.

Additional male children in the household younger than the index child now
reduces only girls' demand for the more expensive service types, private and traditional.
Younger female children also reduce the probability of an index daughter's visit to a
private facility, but by less than do younger males. This lends evidence to the hypothesis
of promale bias for more expensive health care investments.

Preschool girls in residence who are older than the index child have the effect of
increasing the likelihood of boys' public and girls' traditional visits. Adult females age
14-20 raise the demand for son's private care, while females age 21-60 increase the
probability of a daughter's private visit. It is conceivable that adult women have strong
preferences for higher quality child curative investments, and the greater their numbers in
the household, the more ability they may have to exert these preferences. Elderly men in
the household decrease demand for private care for boys only. Although the effects is
negative instead of positive, perhaps adult males in the household inﬂuence male children

more than female children, such as in the ﬁndings of Thomas (1994).

LOW VS. HIGH MATERNAL EDUCATION
The next stratiﬁcation we investigated was maternal education level. The ﬁndings

from the baseline pooled model showed that years of maternal education below the

109

elementary school level had demand effects which differed from those at or above the
level of elementary school completion. We therefore wish to test the hypothesis that
other factors inﬂuence care differently depending on the education level of the mother.
The likelihood ratio test of the hypothesis of no coefﬁcient differences between the group
with less than elementary education and the group with elementary or more education
could be rejected at the 16 percent level.“ Even though this level of statistical
signiﬁcance is not high, will we still proceed with stratifying the analysis by maternal
education to investigate whether mothers with varying education levels respond
differently to health care price, quality, or other household inﬂuences.

Maternal education is hypothesized to raise the technical efﬁciency with which
inputs are used and to increase the allocative efﬁciency of input use through better input
choices for child health (T. W. Schultz, 1964). Often the latter effect is thought to be
most important. For a health care demand model which is not conditioned on current
reported illness, increased technical efﬁciency would imply that long-run demand for
curative care would be reduced because mothers use their chosen health inputs more
efficiently, for example, by providing appropriate nutritional and sanitation behaviors to
complement health care investments. Allocative efﬁciency would imply that mothers
make the better health input choices among the available alternatives. This could include
using health services that are of higher quality when the child is ill. The long-run
reduced-form estimates in this study and hence reﬂect the inﬂuence of both allocative and

technical efﬁciency.

 

8" Chi-square==223, d.f. 203, chiprob 0.160.

110

Education has been hypothesized to affect human capital investments through
increased income, better processing of information, or by changing preferences.
Rosenzweig and Schultz (1982) and Schultz (1984) argue that the beneﬁts of health
infrastructure and the information provided will be greater for the least educated.
Caldwell (1979), on the other hand, hypothesizes that more educated women beneﬁt more
because they are better able to understand information in the media and ﬁom health
personnel. Stratiﬁcations by mother education level will enable us to compare whether
health care demand for children of mothers with little education is more or less
responsive to the availability and quality of health care than is demand for children of
more highly educated mothers.

The effect of education through income will be controlled for in our analysis to a
large degree by including the value of household assets and measures of the extent and
quality of health and community infrastructure in the analysis. Controlling for each of
these factors results in a more pure education effect. Without them, education could
reﬂect household resources if more highly-educated parents also have higher incomes; it
could also pick up more complete and higher quality community infrastructure if better
educated parents live in areas with more and better services.

Other research on health care demand has found mixed results of the effect of
maternal education. In Bicol Philippines, maternal education has no impact on facility
choice for children age 0-5, but increases the demand for private services among the 0-13
age group (Akin et al. 1985). Ching (1995) ﬁnds that education of the household head,
usually a male, reduces demand for modern services for rural and urban Philippine

children age 0-15. Hotchkiss' (1993) results for the determinants of facility for child birth

lll

for Cebu mothers' show that drug availability was a signiﬁcant determinant of choice
only among well-educated women. Mother's education positively inﬂuences the demand
for child clinic visits in rural Peru (Gertler and van der Gaag, 1990) and public and
private care in urban Bolivia (Ii, 1996). It has no impact on demand in Pakistan
(Alderman and Gertler, 1997) or Cote d'Ivoire (Dor and van der Gaag, 1987; Gertler and
van der Gaag, 1990). For Indonesia Deolalikar (1993) ﬁnds no inﬂuence on choice (both
conditional and unconditional on reported morbidity); Gertler et a1. (1995) using a
different Indonesian data set conclude that it increases demand for hospital care and
reduces demand for lower quality nurse care.

Each of the demand studies mentioned above (except that of Deolalikar) uses a
sample which is conditioned on a child illness being reported. Other studies have found
that more highly educated mothers are more likely to report their children as ill (Sindelar
and Thomas, 1991). In the Cebu data, child illness variables are constructed from 24-
hour recall of very detailed morbidity symptoms. This may reduce the degree of
reporting bias in the data. It may still be true, however, that mothers with more education
may be better at recognizing symptoms and are thus more likely to report them. If
reporting behavior is positively inﬂuenced by maternal education, children of more
highly educated mothers may be over-represented in a conditional sample; such a non-
randomly determined sample would result in biased demand estimates. Furthermore, if
more highly-educated mothers also have a greater likelihood of using certain types of
care, this will be an additional source of bias in the maternal education effects. Estimates

which are not conditioned on illness status will not be subject these biases.

112

The results of the unconditional demand model for Cebu infants stratiﬁed by
maternal primary school education are displayed in Table 7.2. We observe for public
services that the effect of distance to that type of care does not vary by whether the
mother has ﬁnished elementary school or not. Distance to private care, on the contrary,
has negative and signiﬁcant effects only for more highly educated mothers. This result
could arise from fact that mothers with higher education are more likely to use private
services in general, so they could be more sensitive to changes in access. Demand for
children of low-education mothers was more responsive distance to traditional services
than children of high-education mothers. Again, this could be because the low-education
group is much more likely to use traditional care and hence to be affected by barriers to
access to it.

The user fee results were notable in that the low-education group had positive and
signiﬁcant demand responses to own-user fees for public and traditional services. This
was not expected; it is conceivable that mothers with little education respond positively to
certain attributes of these lower quality services that remain unmeasured in this analysis,
and these omitted variables are being picked up by the user fee effect. Private user fees
did not inﬂuence the demand for private services for either high- or low-education
mothers.

The high-education mothers had much greater demand responses to structural
quality at facilities. For public care, the only signiﬁcant inﬂuence for low-education
mothers was a positive one for vaccine availability. High-education public demand was
impacted signiﬁcantly and positively by personnel levels, doctor ratios, and ORT

availability, and negatively by the stock of intravenous ORT treatments on hand. It

113

appears, therefore, that policy changes which could greatly beneﬁt low education mothers
would include increasing the availability of child vaccines and reducing distance and
travel time to public clinics. Boosting personnel levels, doctor ratios, and ORT
availability would beneﬁt primarily those who are more highly educated.

Low-education demand for private care was not inﬂuenced at all by private quality;
alternatively, high-education private demand was increased by the availability of
intravenous diarrhea treatments, and decreased by a wider range of family planning
methods offered at private facilities.

Responsiveness of traditional demand to attributes of these providers was stronger
for those with less education. The positive effect on the formal training dummy is larger
in magnitude for mothers who have not completed elementary school.

Individual and household characteristics also varied by maternal education level.
While the male dummy increases private demand for both groups, the effect is much
larger among moms who have not completed elementary school. Demand for traditional
services among the low-educated was positively affected by the child being male.

The ﬁrst piecewise segment of the child age variable (0-6 months) increased
demand for each type of care among the high-education group, but only for private and
traditional care among children of low-education mothers. The second piecewise
segment of child age (7-24 months) decreased the probability of a public and private visit
for both groups, but the lowered the probability of a traditional visit only among the least
educated.

Maternal age had an impact only for post-elementary school mothers and it reduced

their public demand. Height of mother also had inﬂuential effects only among the more

114

highly-educated group; the ﬁrst linear segment increased their public and private demand.
Perhaps these mothers have other human capital or family background characteristics
which boost their preferences for modern health care. The second height spline reduced
high-education demand for public services.

Husband's non-residence for the duration of the two-year survey had negative
inﬂuences on low-education demand for traditional services. The ﬁrst paternal education
spline (years below elementary) had impacts for children of sub-elementary school
mothers, raising demand for public care at the expense of traditional services use.
Advanced paternal schooling had the effect of reducing the likelihood of using traditional
care among high-education group.

Among the low-educated, males younger than the index infant reduced utilization
of traditional services; younger female infants reduced demand private care. For the post-
elementary group, younger males reduced demand for private services, and younger
children of both sexes reduced utilization of traditional care. Elderly men in the
household decreased the demand for the two more expensive types of care, private and

traditional, only for the high-education sample.

LOW VS HIGH ASSET HOUSEHOLDS

Results from the baseline pooled model revealed that asset levels in the ﬁrst 33%
of the distribution inﬂuenced demand differently from values above that level. It is of
interest, therefore, to investigate the possibility that other factors inﬂuence care
differently depending on the asset standing of the child's household. The likelihood ratio

test of the hypothesis of no coefﬁcient differences between the households with asset

115

values only in the ﬁrst tercile and those with values above that level was not signiﬁcant at
any reasonable level of signiﬁcance.87 We will still split the sample by owned asset value
at the ﬁrst tercile, however, to investigate whether children in households with different
resource levels respond differently to health care price, quality, or other household
inﬂuences.

Household resource levels have been shown to have important inﬂuences on
health care choice for children. Moreover, price and quality responses differ by income-
level in some studies. In Bicol, Philippines, Akin et a1. (1985) ﬁnd statistically
signiﬁcant but small effects of household assets on the probability of choosing public and
self-care over traditional care for children. Household income (instrumented using
household non-labor income) has no impact on choice for child health care in Indonesia
(Deolalikar, 1993); it increases the probability of nurse and doctor care relative to self-
care in Cote d'Ivoire (Dor and van der Gaag, 1987).

Higher negative price elasticities for lower-income households (often measured
using household total or per capita consumption levels) are found in the Philippines
(Ching, 1995), Pakistan (Alderman and Gertler, 1997), Indonesia (Gertler et al., 1996),
Peru and Cote d'Ivoire (Gertler and van der Gaag, 1990), and urban Bolivia (Ii, 1996). A
similar result is found for obstetric care choice by Cebu mothers (Hotchkiss, 1993).

Each of the demand studies above (except that of Deolalikar) uses a sample which
is conditioned on a child illness being reported. In the Cebu data, child illness variables
are constructed from 24-hour recall of very detailed morbidity symptoms. This careful

approach collecting morbidity data may reduce the degree of reporting bias in the data. It

 

87 Chi-square=l83, d.f. 203, chiprob 0.840.

116

{/7

may still be true, however, that household with more resources may be more likely to
report an illness episode, for instance if they have had more contact with health providers
in the past and are more familiar with symptoms, or simply because they are better able to
afford treatment. Dow (1995a) estimates the probability of reporting sickness for adults
in Cote d'Ivoire and ﬁnds very large positive inﬂuences for both income and wages. If
reporting behavior is in fact inﬂuenced by household resource levels, higher-income
children may be over represented in a conditional sample; this non-representativeness
would lead to biased demand estimates. Furthermore, if higher-income children also
have a greater likelihood of using certain types of care, this will be an additional source
of bias in the asset effects. Estimates which are not conditioned on illness status will not
be subject these biases.

As in the mother education stratiﬁcations, differences in the distance effects for
these two groups were unexpected. Public demand was decreased much more by distance
for high- rather than low-asset households; the effect is signiﬁcant for both groups, but
the magnitude and the level of signiﬁcance is much higher for the richer stratum. For
private demand, the magnitude of the distance effect was the same for the two income

classes, but was statistically signiﬁcant only for those in the better-off stratum.

 

Traditional distance, on the other hand, is negative and signiﬁcant for both groups, but
the impact is much stronger for the low-income.

User fees were important only for low-asset demand at traditional facilities.
Similar to low-education households, the impact was positive, possibly indicating that
unmeasured desirable characteristics of this service type are correlated with and biasing

the user fee result.

117

Responsiveness to quality at facilities varied by asset standing. Both groups
reacted positively to public vaccine offerings and personnel levels. The current stock of
intravenous diarrhea treatments at public clinics had negative demand impacts for both,
but the magnitude of effect was greater for the poor. As discussed in earlier sections,
because this variable is the result of both supply and demand for drugs, the negative sign
could in fact indicate that this drug is in high demand in public facilities in poor areas
which may be characterized by higher rates and more severe cases of diarrhea. Only
hi gh-income households responded positively to public doctor ratio and ORT availability.

Low-asset private demand was affected negatively by nurse ratio and positively by
supply of intravenous diarrhea treatments at private facilities. The probability of a private
visit for higher-asset children was reduced by current stock of ORT and a wider range of
family planning methods at private facilities. Training of traditional practitioners had
signiﬁcant positive impacts on use of traditional care for both groups, the magnitude of
the effect being greater for the low asset holders.

The impacts of individual and household characteristics also differed by household
asset value. Male gender signiﬁcantly increased the chances of a private visit for both
groups, but the magnitude of the effect was twice as large for the poorer group.
Traditional demand is signiﬁcantly increased by male sex only for the higher stratum.

Mother education increased the likelihood of using private care only among higher-
asset children; however public utilization was increased for the poor by post-elementary
maternal education. Maternal age appears to reduce slightly the probability of public and

traditional visits among higher-asset households. Mother's height has strong positive

118

b.

inﬂuences on demand for public and private care for the low-income, probably capturing
aspects of her human capital or family background not being picked up by her education.
Paternal education reduces demand for traditional services only among high-

resource households. Insurance coverage has positive and similar effects on the
likelihood of a private visit for both asset classes, but the statistical signiﬁcance of the
impact of much greater for the better-off group.

The inﬂuences of household structure varied by household resource status as well.
Males in residence who are younger than the index child and elderly men reduce demand
for the two most expensive type of care, private and traditional, only among better-off
households. It is conceivable that resources are distributed more equally within
households who have low resource levels. Other evidence has indicated that
discrimination against certain types of individuals (e. g., females) is greater at lower
income levels; for example, Alderman and Gertler (1997) ﬁnd that girls' demand for care
is more price elastic than boys' at low income levels, but that the effect diminishes greatly
with income. Perhaps high-income households discriminate more in favor of the very
youngest and very oldest males for private health care than. low—income households do.

Adult females in the household increase the likelihood of a private care visit only
for high-asset children. It is possible that adult females have greater preferences for
better-quality child care and those in higher-asset environments have the access to

resources needed to act more upon these preferences.

CHILDREN 1-6 MONTHS VS. 7-24 MONTHS OF AGE

119

In

In the pooled sample, we saw that demand for services was strongly affected by
child age. From birth to 6 months, age had a positive impact on the likelihood of using
each type of service, whereas in the 7 to 24 month period, demand for modern services
was decreased by child age. These disparate inﬂuences of child age were among the
strongest predictors of health care utilization; furthermore, they arose despite the fact that
illness levels did not fall after the child's sixth month. Given this major reversal in the
effect of age on demand, we could expect to see other determinants of demand differ by
age. One reason these could arise is that health production functions are likely to change
with the stage of infant development. For example, the period between 7 and 24 months
of age may differ from the 0-6 month period because it is when non-breastmilk liquid and
solid nutrients are being introduced. This could make the child much more susceptible to
water- and food-related illnesses such as those which cause diarrhea and anorexia. If
health production technology does vary with age, demands for health inputs and the
determinants of these demands may also. The likelihood ratio test of coefficient
differences between children birth to 6 months and those age 7 to 24 months was not
signiﬁcant.88 We will split the sample by child age at six months, however, because it
will allow us to investigate how household and community factors inﬂuence demand for
health care differently by child age.

Child age has been shown elsewhere to have important impacts on demand for
curative health care services. Alderman and Gertler's results for Pakistan (1997) reveal
that girls' demand for care from government clinics and doctors increases with a linearly

speciﬁed age effect, but boys' does not. In Indonesia, age up to the ﬁfth birthday has

 

3" Chi-square=205, d.f. 203, chiprob 0.447.

120

negative impacts on child visits to doctors and health centers (Gertler et al., 1996);
similar results are found in Bicol, Philippines (Akin et al., 1985) and urban Bolivia (1i,
1996). Deolalikar (1993) also ﬁnds a negative effect of age on unconditional demand for
traditional and modern care in Indonesia. For children age 0-15 in the Philippines, linear
age effects are negative on demand of public and private services (Ching, 1995). In Cote
d'Ivoire, age under ﬁve years has the effect of increasing the demand for lower-cost and
quality nurse care, with the inﬂuence turning negative after ﬁve; age had no impact on
demand for doctor care (Dor and van der Gaag, 1987). Gertler and van der Gaag (1990),
using the same data as Dor and van der Gaag, ﬁnd, on the other hand, that child age zero
to three negatively impacts the demand for hospital and clinic care; age after three has no
effect. The disparity in the ﬁndings on the child age effects using the same data set could
be due to the difference in how the variable is divided into piecewise segments (in the
former the break is at ﬁve years, in the second at three years), or to differences in how the
choice alternatives were deﬁned in each model. Finally, age under three was found to
decrease the likelihood of a visit to a clinic; age over three decreased the probability of a
visit to a private doctor, respectively, in rural Peru (Gertler and van der Gaag, 1990).

As discussed in previous sections, all but one of the health care demand studies
above (Deolalikar) use samples which are conditioned on a current self-reported illness.
If younger children are shown preference for health investments, so their illnesses are
more likely to be reported by the mother than those of lower birth order children, this
could lead to a selected sample for which estimated results would be biased. Models
which are not conditioned on current reported morbidity will not be subject these possible

sources of bias.

121

Table 7.4 presents the health care demand equations stratiﬁed by child age.
Beginning with the facility effects, distance has negative impacts for all types of care for
both age groups. Distance to public facility decreases public demand for younger infants
more than for their older counterparts. For private demand, private distance is a
signiﬁcant deterrent to utilization only for the youngest children. The effect of distance
on demand for traditional services does not vary by child age.

Some user fee inﬂuences also varied by age. Private fees have inﬂuential
negative impacts on private visits only for the youngest; user fees for traditional services
were positive only for older children; public user fees were not important at either age.

A few facility quality results also differed by age. Doctor ratios at public facilities
were positive and important only for older infants. Public supply of intravenous diarrhea
treatments was negative and signiﬁcant for both ages, but the effect was much stronger
for children in the 7-24 month group who are more likely to be weaning.89 Alternatively,
ORT stocks at public facilities had important positive inﬂuences on public demand for
both ages. Public personnel impacts were signiﬁcant and positive, but not large, in both
groups.

For private care, ORT availability had large negative impacts on the youngest
children only, while intravenous diarrhea treatments had positive and important effects
only for the older group. It was also found that the range of available family planning

methods at private facilities signiﬁcantly reduced only younger children's demand.

 

8" This again highlights the notion that if public facilities run low on supplies
because children in the area have high rates diarrhea, low supply could be related to very
high demand for the commodity.

122

The two measured aspects of traditional service quality, education and formal
training, raised the probability of a traditional visit only among children older than six
months of age.

Individual- and household-level characteristics also varied by child age. The
effect of child being male had positive impacts on private and traditional services for all
ages; however, the magnitude of this effect for newborns was double what is was for the
older infants. Male gender raised public demand only for children younger than six
months.

The ﬁrst piecewise segment of maternal education (0-5 years) was positive and
signiﬁcant for pre-six month demand for pubic and traditional care, and for post-six
month demand for private services. Maternal education beyond ﬁve years raised the
likelihood of a private visit for post-six month olds; it reduced the chances of a traditional
visit for those younger than six months of age. The ﬁrst mother height spline had
positive effects on using private services for all ages. Public demand for older infants
was increased by height in the ﬁrst spline, but decreased by height in the second spline.
Child's mother being the senior woman in the household boosted the chances of a
newborn visit to high-quality, high-priced private care.

Husband education up to the ﬁfth year had the effect of increasing newborn
demand for public care; the impact of his schooling after the ﬁfth year was to raise the
probability of a newbom's visit to a private facility reduce the probability of a newborn
being taken to a lower-quality facility.

Also interesting were the insurance and asset effects. Insurance coverage

increased the likelihood of a private visit for only older infants. Asset values in the lower

123

33 percent of the distribution had large and positive inﬂuences on private demand for
both ages; values above the 33rd percentile, however, did not signiﬁcantly affect demand
for private care, but served to reduce the likelihood of a public visit and increase that of a
traditional visit among 7-24 olds.

Household composition effects had varying inﬂuences depending on child age.
Additional male children who are younger than the sample infant (these would not be
siblings, but other infants in residence), reduced the chances of 7-24 month olds visiting
private or traditional service providers, which are two higher-priced alternatives. In
contrast, preschool girls older than the index child increased the likelihood of a traditional
visit for non-newborns. Young adult females raised the demand for private care for all
infants, while the presence of elderly men negatively affected private and traditional care

only for 7-24 month olds.

CONCLUSION

Differential responses to health facility characteristics, household resources and
composition were found when the demand analyses were stratiﬁed by demographic and
socioeconomic subgroup. Starting with the care attributes, distance to public facilities
had greater negative impacts on public visits for girls than boys, for higher-asset than
lower-asset households, and for newborns than older infants. Distance to private facilities
reduced the chances of a private visit more for males, for the high-asset and high-
education groups, and for newborns. Traditional distance dampened traditional demand
most among daughters, low-education mothers, and low-asset households. The ﬁnding

that greater asset households responded more to distance to modern care was somewhat

124

 

surprising since the other research has found that the poor are more sensitive to distance
to facilities. However, past studies have not allowed differential responses to distance by
type of care being sought. If these effects had been constrained to equality, it is very
likely that the conclusion would have been that lower-income households are more
responsive to distance; this is because the low-asset coefﬁcient on traditional distance is
much larger than any of the high-asset distance effects and it would have dominated the
constrained effects.

Differences in user fee responses were less varied. Less-educated mothers had
positive responses to public user fees. Newborn private care was inhibited by private
user fees. Demand for traditional services among girls, less-educated mothers, low-asset
households, and older infants were increased by traditional user fee. In general the
positive price effects were not expected; however, because there is not much data
available to control for the quality attributes which drive the demand for traditional
services, it is possible that the positive inﬂuence is picking up unmeasured desirable
aspects of the quality of this type of care. Furthermore, it is conceivable that poor
mothers with little education are most attracted to traditional providers for reasons we are
not aware of . Moreover, since girls and older infants appear to be discriminated against
for the use of public services, their demand for traditional care may be more price
inelastic than that of boys and newborns.

Results for the facility quality effects were not as straightforward. For public
quality responses did not differ by child sex; mothers with more education responded

more positively to public quality than did the less-educated.

125

Private quality inﬂuences were stronger for girls, but the direction of the effects
was not uniform. Mothers with more education had only slightly stronger to responses to
private quality. The asset and age stratiﬁcations yielded not clear set of private quality
impacts.

For traditional care, again girls, non-newboms, and mothers with little education
had strong positive responses to provider characteristics.

Individual and household effects also varied with some of the stratiﬁcations. The
positive inﬂuence of male gender on private demand was strongest among newborns,
low-asset households, and mothers with less education. The male inﬂuence on traditional
visits was paramount for newborns, high-asset households, and low-educated mothers.

The ﬁrst linear segment of maternal education raised the chances of a public visit
among the poor and newborns; it enhanced the chances of a private visit for boys, the less
poor and non-newboms; and it strengthened traditional demand for girls and newborns.
Maternal education over ﬁve years reduced private demand for girls but increased it for
non-newborns. Maternal height raised the chances of a modern visit for children of more
highly educated mothers and with low-asset holdings. Paternal education raised the
demand for modern services for newborns.

The presence of children younger than the index infant reduced private and
traditional demand for girls and children in high-asset households most. Elderly men, on
the other hand, reduced the likelihood of private and traditional visits for boys, non-

newboms, and infants of more highly educated mothers.

126

 

hr
bc

fa.

Section 8

DISCUSSION AND CONCLUSIONS

This research seeks to add to the literature on child health care demand in
developing countries in several ways. First, previous research has focused mainly on
care for adults; this is one of the few studies to examine demand for services to treat
young children. This is an important line of inquiry as the ﬁrst three years of life is the
most crucial period in terms of physical and mental development (Martorell, 1995).
Illness during this period, much of which is due to treatable infectious disease, can have
devastating effects on development because feeding, appetite, and absorption of nutrients
can be severely reduced (Adair, et al., 1993).

Second, in addition to information collected from a broad, multi-wave survey of
households, the study uses a unique set of data on quality of health facilities and
characteristics of health personnel. This provides the opportunity to examine how
speciﬁc aspects of services drive demand.

Third, the study uses BOTH distance and fees for service as measures of price.
Most studies in this genre use only distance to estimate price responses; policy
recommendations for user fees are often made based on these results. However, if
households respond differently to changes in distance than to changes in fees, this could
be important for predicting the effects of changing user fees versus building new

facilities.

127

f0

Fourth, traditional health providers, which are used very frequently in developing
countries but which are not always considered in demand studies, are included in the set
of modeled health care alternatives.

Fiﬁh, the empirical approach allows differential price and quality responses by
type of health care. Most studies constrain price and quality coefficients to be equal
across health care alternatives and assume that the same set of health care characteristics
impact the probability of visiting different types of providers. This greatly limits the
insights of what inﬂuences improving quality or raising fees at different types of facilities
will have on demand.

Sixth, and ﬁnally, the model provides estimates of price and income responses
that are not conditioned on self-reported morbidity status; this approach avoids the
selection bias inherent in using a sample of only those who report themselves as ill. A
strong argument can be made that self-reported health measures are endogenous in a
model of health care demand because of unobserved attitudes that may affect both health
reporting and utilization when an illness occurs, and should therefore, not be conditioned

on.

RESULTS

Results from our baseline ﬂexible MNL model indicate that when differential
inﬂuences of price and quality on demand are allowed to vary by the type of care, there
are indeed differential demand responses by service type. These results change
dramatically when provider attribute effects from these diverse types of services are

forced to have the same set of elements describing them, and each element is forced to

128

have an identical impact on the demand for each type of care. Forcing different ﬂavors of
health care to be inﬂuenced by the same set of attributes, and imposing the restriction that
each of these attributes has the same effect on every kind of service, is unrealistic at best
and could be highly misleading.

We ﬁnd that distance to facilities has consistently strong negative effects, and it
appears from these estimates that distance is a more important determinant of demand
than monetary user fees. In the context of health service reforms in developing countries,
this demonstrates that location of facilities may largely determine not only whether health
care is sought, but the type of care chosen.

User fee variables were very sensitive to whether non-facility community effects
were controlled for. This was especially true for the public fee variable. Fees have
strong negative impacts on demand for public care when other community attributes are
either replaced by district-level dummies or omitted altogether; their impacts, however,
are close to zero and insigniﬁcant when these inﬂuences ARE controlled for. This is an
important ﬁnding since results from these types of studies may be used to set pricing
policies for public health care in developing countries: without the detailed community
controls we would conclude demand is price sensitive, whereas with them we would
come to a very different conclusion. Past studies have controlled for few, if any, non-
facility community variables. Because the presence and quality of health facilities is
expected to be highly correlated with other community characteristics, facility price and
quality results from models where community attributes are not speciﬁcally accounted for

should be interpreted with caution.

129

Strong, though differential, quality effects were found on demand for all facilities.
Availability of ORT, vaccines, family planning services, and personnel had large positive
effects on the probability of using public services; more sophisticated (intravenous)
diarrhea treatments increased visits to private facilities; demand for traditional providers
was increased by the practitioner having had any formal health training. These particular
quality effects were generally robust to the nesting speciﬁcation.” However, similar to
the public price variable, many of the quality variables were sensitive to whether other
community characteristics were controlled for. Furthermore, a few of the quality
variables, mainly current drug supplies at the facility, had consistently unexpected
negative impacts on demand. This result could be driven however by the fact that the
drug variables were deﬁned as those currently in stock. The availability of a drug at any
particular time reﬂects the interaction of both supply and demand factors; if a certain drug
is out of supply it may be due to excess demand for it. For instance, if public clinics with
intravenous drugs are in areas with high demand-poorer areas with high diarrhea levels--
their stock may be diminished quickly; this would produce a negative correlation between
demand for services and availability of drugs (Mwabu, et al., 1993). The signiﬁcant
negative effect of intravenous diarrhea treatments may, therefore, indicate that they are

high demand at public health facilities in high-diarrhea areas.

 

9° Cross-effects between public and traditional services were also found in the

nested models. Increasing barriers to access (distance or user fee) at one raises
demand at the other type of facility. This signals that public facilities could in
principle be major channels for the delivery of efﬁcacious modern health services in
developing countries if their access were expanded.

130

Another possiblity is that effect arises from non-random placement of facilities.
While formal tests of the hypothesis cannot be performed because the complete set of
facility quality data is not available at two different points in time (Pitt, Rosenzweig,
Gibbons, 1995), an informal test was performed to get an indication of how facility
quality was related to community illness rates. The complete set of regressors from the
demand models was used to estimate a logit model of any reported illness using the 24
hour recall data. Several sets of facility quality variables had positive eﬂects on reported
illness. Facilities may therefore be purposefully located in sicker areas; this endogenous
placement of quality of services makes their true effect on demand difﬁcult to assess.9|

For the individual- and household-level variables, we ﬁnd that utilization of
modern services is increased by household income and maternal human capital. Maternal
education and height increases demand for both public and private, but especially private
care. As incomes and health insurance coverage rise, households switch from public and
traditional care into the private services. This income bifurcation in where people seek
care could have equity considerations which are policy relevant: it implies that investing

resources in public facilities may be a good method of targeting health care resources to

 

9' Mother age, household asset values, and to some extent mother education
reduced the probability of reported illness.

Public distance and price increased as one would expect. Five of the seven public
quality variables, however, had positive effects on reported illness: these were doctor
ratio and nurse ratio, and ORT, other diarrhea drug, and vaccine availability. While none
these was individually signiﬁcant, their joint effect was. For private facilities, price was
positive; however, distance had a negative effect, and number of personnel, doctor and
nurse ratio each had signiﬁcant positive effects on reported illness. These results may
indicate that facilities may be placed in sicker areas.

131

 

the poor because they are the primary users of these services.92 Similar results were
found in Jamaica by Gertler and Sturm (1997).

Demand is greater among children who are male and less than 6 months of age;
these groups are also more likely to be taken to expensive, higher quality private
facilities. The presence of infant males younger than the index child reduces the
likelihood of an index child's visit, especially for females. These individual and
household effects were diminished somewhat in the nested model where correlation is
allowed across facility types.

Differential responses to health facility characteristics, household resources and
composition were found when the demand analyses were stratiﬁed by demographic and
socioeconomic subgroup. Distance to public facilities had greater negative impacts on
public visits for girls than boys, for higher-asset than lower-asset households, and for
newborns than older infants. Distance to private facilities reduced the chances of a
private visit more for males, for the high-asset and high-education groups, and for
newborns. Traditional distance dampened traditional demand most among daughters,
low-education mothers, and low-asset households. The ﬁnding that greater asset
households responded more to distance to modern care was somewhat surprising since
the other research has found that the poor are more sensitive to distance to facilities.
However, past studies have not allowed differential responses to distance by type of care

being sought. If these effects had been constrained to equality, it is very likely that the

 

92 This raises possible concerns about the development of a two-tiered system in

the long-run. However, given the current organization of service delivery, this may
have important welfare effects.

132

conclusion would have been that lower-income households are more responsive to
distance; this is because the low-asset coefﬁcient on traditional distance is much larger
than any of the high-asset distance effects and it would have dominated the constrained
effects.

Differences in user fee responses were less varied. Less-educated mothers had
positive responses to public user fees. Newborn private care was inhibited by private
user fees. Demand for traditional services among girls, less-educated mothers, low-asset
households, and older infants were increased by traditional user fee. In general the
positive price effects were not expected; however, because there is not much data
available to control for the quality attributes which drive the demand for traditional
services, it is possible that the positive inﬂuence is picking up unmeasured desirable
aspects of the quality of this type of care. Furthermore, it is conceivable that poor
mothers with little education are most attracted to traditional providers for reasons we are
not aware of . Moreover, since girls and older infants appear to be discriminated against
for the use of public services, their demand for traditional care may be more price
inelastic than that of boys and newborns.

Results for the facility quality effects were not as straightforward. For public
quality responses did not differ by child sex; mothers with more education responded
more positively to public quality than did the less-educated. Private quality inﬂuences
were stronger for girls, but the direction of the effects was not uniform. Mothers with
more education had only slightly stronger to responses to private quality. The asset and

age stratiﬁcations yielded not clear set of private quality impacts. For traditional care,

133

again girls, non-newborns, and mothers with little education had strong positive
responses to provider characteristics.

Individual and household effects also varied with some of the stratiﬁcations. The
positive inﬂuence of male gender on private demand was strongest among newborns,
low-asset households, and mothers with less education. The male inﬂuence on traditional
visits was paramount for newborns, high-asset households, and low-educated mothers.

The ﬁrst linear segment of maternal education raised the chances of a public visit
among the poor and newborns; it enhanced the chances of a private visit for boys, the less
poor and non-newborns; and it strengthened traditional demand for girls and newborns.
Maternal education over ﬁve years reduced private demand for girls but increased it for
non-newborns. Maternal height raised the chances of a modern visit for children of more
highly educated mothers and with low-asset holdings. Paternal education raised the
demand for modern services for newborns.

The presence of children younger than the index infant reduced private and
traditional demand for girls and children in high-asset households most. Elderly men, on
the other hand, reduced the likelihood of private and traditional visits for boys, non-

newboms, and infants of more highly educated mothers.

134

 

BIBLIOGRAPHY

135

st :11. L11.- Fuw

 

Ira-m .

BIBLIOGRAPHY

Acton, Jan. 1975. Nonmonetary Factors in the Demand for Medical Service: Some
Empirical Evidence, Journal of Political Economy, 83.3:595-614.

Adair L., B. Popkin, J. VanDerslice, J. Akin, D. Guilkey, R. Black, J. Briscoe, W.
Flieger. 1993. Growth dynamics during the ﬁrst two years of life: A Prospective Study
in the Philippines, European Journal of Clinical Nutrition, 47.1 :42-51.

Akin, John S., Charles Grifﬁn, David Guilkey and Barry Popkin. 1986. The Demand for
Primary Health Care Services in the Bicol Region of the Philippines, Economic
Development and Cultural Change, 34.4:755-782.

Akin, John S., David Guilkey, Charles Grifﬁn and Barry Popkin. 1985. The Demand for
Primary Health Services in the Third World. Totowa, N.J., U.S.A.: Rowman and
Allenheld.

Alderman, Harold, and Paul Gertler. 1997. Family Resources and Gender Differences in
Human Capital Investments: The Demand for Children's Medical Care in Pakistan,
Intrahousehold Resource Allocation: Methods, Models, and Policy, ed. Lawrence
Haddad, John Hoddinott, and Harold Alderman. Baltimore: Johns Hopkins University
Press.

Alderman, Harold, and Paul Gertler. 1989. The Substitutability of Public and Private
Health Care for the Treatment of Children in Pakistan, Living Standards Measurement
Study No. 57, World Bank, Washington, DC.

Alderman, Harold and Victor Lavy. 1996. Household Responses to Public Health
Services: Cost and Quality Tradeoffs, The World Bank Research Observer, 11.1:3-22.

Alderman, H., and C. H. Paxson. 1992. Do the poor insure? A synthesis of the
literature on risk and consumption in developing countries. Discussion Paper 164.
Princeton, N.J., U.S.A.: Princeton University, Woodrow Wilson School of Public and
International Affairs.

 

{m

Aligaen, M., and C. F lorencio. 1980. Intrahousehold nutrient distribution and adequacy
of food and nutrient intake of Filipino urban households. Philippine Journal of Nutrition

33 (1): 11-19.

Amemiya, Tekashi. 1978. On a Two-Step Estimation of a Multivariate Logit Model,
Journal of Econometrics 8: 1 3-21.

136

Appleton, Simon, John Hoddinott and John MacKinnon. 1996. Education and Health in
Sub-Saharan Aﬁ'ica, Journal of International DeveIOpment 8.3: 307-39.

Barrera, A. 1990. The role of maternal schooling and its interaction with public health
programs in child health production. Journal of Development Economics 32 (1): 69-91.

Behrman, J ., and A. Deolalikar. 1988. Health and nutrition. In Handbook of
development economics, vol. 1, ed. H. Chenery and T. N. Srinivasan. Amsterdam and
New York: Elsevier.

Behrman, J. R., and A. B. Deolalikar. 1989. Seasonal demands for nutrient intakes and
health status in rural south India. In Seasonal variability in Third World agriculture, ed.
D. E. Sahn, 66-78. Baltimore, Md., U.S.A.: Johns Hopkins University Press for the
International Food Policy Research Institute.

Behrman, J. R., and A. B. Deolalikar. 1989. Wages and labor supply in rural India: The
role of health, nutrition, and seasonality. In Seasonal variability in Third World
Agriculture: The consequences for food security, ed. D. E. Sahn, 107-118. Baltimore,
Md., U.S.A.: Johns Hopkins University Press for the International Food Policy Research
Institute.

Behrman, J. R., and B. L. Wolfe. 1987. How does mother's schooling affect the family's
health, nutrition, medical care usage, and household sanitation? Journal of Econometrics
36: 185-204.

Ben-Akiva, Moshe and Steven R. Lerman. 1985. Models of Multidimensional Choice
and the Nested Logit Model. Discrete Choice Analysis: Theory and Application to
Travel Demand, Cambridge: MIT Press.

Benefo, Koﬁ, and T. Paul Schultz. 1994. Determinants of Fertility and Child Mortality
in Cote d'Ivoire and Ghana, Living Standards Measurement Study No. 103, World Bank,
Washington, DC.

Birdsall, Nancy and Punham Chuhan. 1986. Client Choice of Health Treatment in Rural
Mali, World Bank, Population and Human Resources Department, Washington, DC.
Processed.

Bouis, Howarth, Marilou Palabrica-Costello, Orville Solon, Daniel Westbrook, Azucena
B. Limbo. 1997. Investing in Adolescents in Rural Household: A Philippine Case
Study. Manuscript. International Food Policy Research Institute.

Bouis, H., and C. Pena. 1997. Inequality in the intrafarnily distribution of food: The

dilemma of deﬁning an individual's "fair share." In Intrahousehold resource allocation:
Methods, application, and policy, ed. L. Haddad, J. Hoddinott, and H. Alderman.

137

Baltimore, Md., U.S.A.: Johns Hopkins University Press for the International Food
Policy Research Institute.

Bouldac, Dennis, Guy Lacroix and Christophe Muller. 1996. The Choice of Medical
Care Providers in Rural Benin: A Comparison of Discrete Choice Models. Journal of
Health Economics 15: 477-98.

Butler, J. S., R. V. Burkhauser, J. M. Mitchell, and T. P. Pincus. 1987. Measurement
error in self-reported health variables. Review of Economics and Statistics LXIX (4):
644-650.

Cebu Study Team. 1992. A child health production function estimated from longitudinal
data. Journal of Development Economics 38 (2): 323—35 1.

Chambers, R., R. Longhurst, and A. Pacey. 1981. Seasonal dimensions to rural poverty.
London: Frances Pinter.

Chen, L.C., A. Chowdhury, S.L. Huffman. 1980. Anthropometric Assessment of
Energy-protein Malnutrition and Subsequent Risk of Mortality among Preschool Aged
Children, American Journal of Clinical Nutrution, 33.8: 1836-45.

Chen, L. C., E. Huq, and S. d'Souza. 1981. Sex bias in the family allocation of food and
health care in rural Bangladesh. Population and Development Review 7 (1): 55-70.

Collins, David H., J. Quick, S.N. Musau and D. L. Kraushaar. 1995. Health Financing
Reform in Kenya: The Fall and Rise of Cost Sharing. 1 989-94. Management Sciences
for Health, Boston, MA, Stubbs Monograph Series.

Corrria, Giovanni and Germano Mwabu. 1997. UNU/WIDER Working Papers No.

Corrria, Giovanni, Richard Jolly and Frances Steward (Eds). 1987. Adjustment with a
Human F ace, Vol. 1, Oxford: Clarendon Press.

Creese, Andrew. 1991. User Charges for Health Care: A Review of Recent Experience,
Health Policy and Planning, 6.4:309-319.

Creese, Andrew and J. Kutzin. 1995. Lessons From Cost Recovery in Health. World
Health Organization. National Health Systems and Policies Unit, Discussion Paper 2,
Geneva, Switzerland.

Deaton, Angus. 1988. Quality, quantity, and spatial variation in price, American
Economic Review, 78.3: 418-430.

De Ferranti, D. 1985. Paying for Health Services in Developing Countries: An
Overview, World Bank Staff Working Papers No. 721.

138

Deolalikar, Anil. 1993. The Demand for Health Services in Indonesia: The Role of
Prices, Service Quality, and Reporting of Illnesses, Institute for Economic Research,
Discussion Series #93-13, University of Washington.

Donabedian, Avedis. 1988. The Quality of Care: How Can It Be Assessed?, Journal of
the American Medical Association, 260.12: 1743-1748.

Donabedian, Avedis. 1980. Volumes 1 and 2: The deﬁnition of quality and approaches
to its assessment and monitoring, Health Administration Press, Ann Arbor, p. 1-163, 3-
383..

Dor, Avi, P. Gertler, and Jacques van der Gaag. 1993. Non-price rationing and the
medical provider choice in rural Cote d’Ivoire, Journal of Health Economics, 6.4: 291-
304.

Dor, Avi, and Jacques van der Gaag. 1987. The Demand for Medical Care in
Developing Countries: Quantity Rationing in Rural Cote d’Ivoire, Living Standards
Measurement Study No. 35, World Bank, Washington, DC.

Dow, William H. 1995a. Unconditional Demand for Curative Health Inputs: Does

Selection on Health Status Matter in the Long Run? Rand Labor and Population Program
Working Paper Series 95-22, Dru-1234-rc.

Dow, William H. 1995b. Discrete Choice Estimation of Price-elasticities: The Beneﬁts
of a Flexible Behavioural Model of Health Care Demand, Rand Labor and Population
Program Working Paper Series 95-20, Dru-1232-rc.

Ellis, Randall and Germano Mwabu. 1991. The demand for rural outpatient care in rural
Kenya, mimeo, Dept. of Econorrrics, Boston University.

Evenson, R. E., B. M. Popkin, and E. King-Quizon. 1980. Nutrition, work, and
demographic behavior in rural Philippine households: A synopsis of several Laguna
household studies. In Rural household studies in Asia, ed. H. P. Binswanger, R. E.
Evanson, C. A. Florencio, and B. N. F. White. Singapore: Singapore University Press.

Filmer, Deon and Lant Pritchett. 1997. The impact of health spending on health
outcomes. World Bank Research Report.

Foster, Andrew. 1992. Rice prices, credit markets and child growth in rural Bangladesh,
mimeo, Dept. of Economics, University of Pennsylvania.

Frankenberg, Elizabeth. 1995. The Effects of Access to Health Care on Infant Mortality
in Indonesia, Health Transitions Review, 5 (2, November).

139

Garner, Paul, Jane Thompson and Daryl Donaldson. 1990. Quality Assessment of
Health Facilities in Rural Papua New Guinea, Health Policy and Planning, 51.1 :49-59.

Gertler, Paul and Elaina Rose. 1997. Collecting Health Data in Developing Countries.
LSMS draft chapter.

Gertler, Paul, L. Locay and W. Sanderson. 1987. Are User Fees Regressive? The
Welfare Implications of Health Care Financing Proposals in Peru, Journal of
Econometrics, 36 (Supp.):67-88.

Gertler, Paul, Jack Molyneaux, Carl Sennato, S. Ahmed, S. Intrajaya, and Elizabeth
Frankenberg. 1995. Pricing Public Health Care Services: Results from a Social
Experiment in Indonesia, Rand Corporation, Santa Monica, Ca. Processed.

Gertler, Paul and Roland Sturm. 1997. Private Health Insurance and Public
Expenditures in Jamaica, Journal of Econometrics, 77:237-257.

Gertler, Paul and Jacques Van Der Gaag. 1990. The Willingness to Pay for Medical
Care: Evidence from Two Developing Countries, Baltimore: Johns Hopkins Press.

Glewwe, Paul, Elizabeth King, Linda Adair and Hanan Jacoby. 1994. The Impact of
Child Health on Educational Outcomes in a Developing Country: An Analysis of
Longitudinal Data from Cebu, Philippines. World Bank, PRDPH, Washington, DC.

Processed.
Greene, William. 1996. Nested Logit Models. Limdep Version 7.0 Manual.

Griffin, Charles. 1992. Health Care in Asia: A Comparative Study of Cost and
Financing, Washington, DC. The World Bank.

Griffin, Charles. 1992. User Charges for Health Care in Principle and Practice, World
Bank Economic Development Institute Seminar Paper No. 37, Washington DC.

Grossman, Michael. 1972. On the Concept of Health Capital and the Demand for
Health, Journal of Political Economy, 80.2:223-255.

Haddad, L., R. Kanbur, and H. Bouis. 1993. Intrahousehold inequality and average
household well-being: Evidence on calorie intakes and energy expenditures from the
Philippines. International Food Policy Research Institute, Washington, DC. Mimeo.

Haddad, Lawrence, Christine Pena, Chizuru Nishida, Agnes Quisumbing, and Alison
Slack. 1996. Food Security and Nutrition Implications of Intrahousehold Bias: A
Review of Literature, Food Consumption and Nutrition Division (FCND) Discussion
Paper No. 19, Washington, DC: International Food Policy Research Institute.

140

 

Hamilton, Lawrence. 1992. Regression with Graphics, Paciﬁc Grove, Ca: Brooks/Cole
Publishing Company.

Heller, Peter. 1981. A Model of Demand for Medical and Health Services in Peninsular
Malaysia, Social Science and Medicine, 16:267-284.

Hensher, David A. 1986. Sequential and Full Information Maximrun Likelihood
Estimation of a Nested Logit Model. The Review of Economics and Statistics.

Herrin, Alejendro. 1992. Micro Impacts of Macroeconomic Adjustment Policies on
Health, Nutrition and Education, Working Paper Series No. 92-18, Manila, Philippines:
Philippine Institute for Development Studies.

Herrin, Alejendro. 1990. An Assessment of Population, Health and Education Policies
in the Philippines, 1986-1988, Working Paper Series No. 90-10, Manila, Philippines:
Philippine Institute for Development Studies.

Hoffman, Saul D. and Greg J. Duncan. 1988. Multinomial and Conditional Logit
Discrete-Choice Models in Demography. Demography, 25:415-427.

Horton, S. 1988. Birth order and child nutritional status: Evidence from the Philippines.
Economic Development and Cultural Change 36 (2): 341—3 54.

Hotchkiss, David. 1993. The Role of Quality in the Demand for Health Care: Cebu,
Philippines, HFS Small Applied Research Paper No. 12, Abt Associates Inc., Bethesda,
Md.

Ii, Mosaka. 1996. Demand for Child Health Care in Urban Bolivia. LSMS Working
Paper.

Jacoby, Hanan and Emmanuel Skouﬁas. 1992. Risk, Seasonaility and School
Attendance: Evidence from Rural India, Rochester Center for Economic Research
Working Paper no. 328. Dept. of Economics, University of Rochester.

Kutzin, Joseph. 1994. Experience with Organizational and Financing Reform of the
Health Sector. World Health Organization, Division of Strengthening Health Services,
Geneva, Switzerland.

Lavy, Victor and Jean-Marc Germain. 1994. Quality and Cost in Health Care Choice in
Developing Countries, Living Standards Measurement Study No. 105, World Bank,
Washington, DC.

Lavy, Victor, John Strauss, Duncan Thomas, and de Vreyer. 1996. Quality of Health

Care, Survival and Health Outcomes in Ghana, Journal of Health Economics, 15.3: 333-
57.

141

 

Livtack, Jennie, and Claude Bodart. 1993. User Fees and Improved Quality of Health
Care Equals Improved Access: Results of a Field Experiment in Cameroon, Social
Science and Medicine, 37.3:369-3 83.

McFadden, Daniel. 1981. Econometric Models of Probabilistic Choice, in C. Manski
and D. McFadden (Eds), Structural Analysis of Discrete Data: With Econometric
Applications, Cambridge, MA: MIT Press.

Martorell, R. 1995. Results and Implications of the Incap Follow-up Study, Journal of
Nutrition, 125.4: 1127S-1138S.

Murdoch, Jonathan. 1990. Risk, production and saving: Theory and evidence from
Indian households, mimeo, Dept. of Economics, Harvard University.

Murdoch, Jonathan. 1991. Consumption Smoothing Across Space: Tests for Village-
Level Responses to Risk, mimeo, Dept. of Economics, Harvard University.

Mwabu, Germano. 1997. User Charges for Health Care: A Review of the Underlying
Theory and Assumptions. UNU/WIDER Discussion Papers No. 127.

Mwabu, Germano. 1989. Nonmonetary Factors in the Household Choice of Medical
Facilities, Economic Development and Cultural Change, 37.2:3 83-392.

Mwabu, Germano. 1986. Health Care Decisions at the Household Level: Results of a
Rural Health Survey in Kenya, Social Science and Medicine, 22.3:315-319.

Mwabu, Germano, Martha Ainsworth and Andrew Nyamete. 1993. Quality of Medical
Care and the Choice of Medical Treatment in Kenya: An Empirical Analysis, Journal of
Human Resources, 28.4:838-862.

Mwabu, Germano and Joseph Wang'ombe. Health Service Pricing Reforms in Kenya,
International Journal of Social Economics, forthcoming.

Panis, Constantijn. 1994. The Piecewise Linear Spline Transformation. Stata Technical
Bulletin, 1 8 :27-29.

 

Park, Changwon and Ben Senauer. 1996. Estimation of Household Brand-Size Choice
Models for Spaghetti Products with Scanner Data. The Retail Food Industry Center,
Working Paper 96-01. University of Minnesota.

Peabody, John W., Omar Rahman, Kristen Fox, and Paul Gertler. 1994. Quality of Care
in Public and Private Primary Health Care Facilities: Structural Comparisons in Jamaica,

Bulletin of the Pan-American Health Organization, 28.2: 122-141.

Ravallion, Martin. 1987. Markets and Famines, Oxford: Clarendon Press.

142

Razzaque, Abdur, Nurul Alarn, Lokky Wai, Andrew Foster. 1990. Sustained Effects of
the 1974-75 famine on infant and child mortality in a rural area of Bangladesh,
Population Studies, 44.1: 145-154.

Reddy, Sanjay and Jan Vandemoortele. 1996. User Financing of Basic Social Services:
A Review of Theoretical Arguments and Empirical Evidence. UNICEF, EPP Working
Paper. New York.

Rosenzweig, M., and T. P. Schultz. 1983. Estimating a household production function:
Heterogeneity, the demand for health inputs and their effects on birth weight. Journal of
Political Economy 91 (5): 723—746.

Rosenzweig, M., and O. Stark. 1989. Consumption smoothing, migration, and marriage:
Evidence from rural India. Journal of Political Economy 97 (4): 905—927.

Ruel M., J. Rivera, J. Habicht, R. Martorell. 1995. Differential Response to Early
Nutrition Supplementation: Long-term Effects on Height at Adolescence, International
Journal of Epidemiology 24.2: 404-12.

Sahn, D. E., ed. 1989. Seasonal variability in Third World agriculture: The
consequences for food security. Baltimore, Md., U.S.A., and London: Johns Hopkins
University Press for the International Food Policy Research Institute.

Sauerbom, R., A. Nougtara and E. Latimer. 1994. The Elasticity of Demand for Health

Care in Burkina F aso: Differences Across Age and Income Groups, Health Policy and
Planning 9.2: 185-92.

Schultz, P., and A. Tansel. 1995. Measurement of returns to adult health: Morbidity
eﬁfects on wage rates in Cdte d'Ivoire and Ghana. LSMS Working Paper, World Bank,
Washington, DC.

Schultz, T. Paul and AysitTansel. 1992. Measurement of returns to adult health:
Morbidity effects on wage rates in Cote d’Ivoire and Ghana, Economic Growth Center
Discussion Paper No. 663, Yale University.

Schwartz, J ., John Akin, and Barry Popkin. 1988. Price and Income Elasticities of
Demand for Health Care: The Case of the Infant Delivery in the Philippines, The World
Bank Economic Review, 2.1 :49-76.

Sindelar, J ., and D. Thomas. 1991. Measurement of Child Health: Maternal Response
Bias. Economic Growth Center Paper No. 663. Yale University.

Strauss, J. 1990. Households, commrmities, and preschool children's nutrition outcomes:
Evidence from rural Cote d'Ivoire. Economic Development and Cultural Change 38 (2):
23 1-262.

143

Strauss, John, Paul Gertler, Omar Rahman and Kristen Fox. 1993. Gender and Life-
Cycle Differentials in the Patterns and Determinants of Adult Health, Journal of Human
Resources, 28.4.

Strauss, John and Duncan Thomas. 1995. Human Resources: Empirical Modeling of
Household and Family Decisions, Handbook of Development Economics 3, ed. T. N.
Srinivasan and J. Behrman. Amsterdam: North Holland.

Strauss, Thomas, and de Vreyer (1996)

Thomas, Duncan, Victor Lavy, and John Strauss. 1996. Public Policy and
Anthropometric Outcomes in Cote d'Ivoire, Journal of Public Economics,61.2: 155—192.

Waddington, C. and K. Enyimayew. 1989. A Price to Pay: The Impact of User Charges
in Ashanti-Akim District of Ghana, International Journal of Health Planning and
Management, 4:17-47.

Wong, Emelita, Barry Popkin, David Guilkey, and John Akin. 1987. Accessibility,
Quality of Care and Prenatal Care Use in the Philippines, Social Science and Medicine,
24.1 1:927-944.

World Bank. 1987. Financing Health Services in Developing Countries: an Agenda for
Reform, A World Bank Policy Study, Washington, DC.

Yoder, R. 1989. Are People Willing and Able to Pay for Health Services?, Social
Science and Medicine, 29:35-42.

144

APPENDICES

145

APPENDIX A

Main Tables

146

 

Table 4.1 Quality Attributes by Disaggregated Facility Type

 

 

 

Variable bggyhs pubcln pubhosp prvcln prvhosp Total
CthutpTrtd 1.00 1.00 1.00 0.86 1.00 0.86
ChdImmunAvl 1.00 0.96 0.75 0.57 0.67 0.87
PrenUserFee 0.00 0.00 0.00 2.88 3.59 0.97
ChOutpUsFee 0.00 0.00 0.00 7.76 9.84 2.55
# Personnel 10.52 25.19 148.71 6.75 96.55 41.01
Doctr Ratio 0.10 0.10 0.27 0.54 0.21 0.18
Nuts Ratio 0.69 0.48 0.55 0.31 0.57 0.54
ORT CunAvl 0.72 0.58 0.29 0.00 0.50 0.52
InterurAvl 0.03 0.06 0.43 0.25 0.60 0.20
VaccinIndex 0.05 0.68 0.44 0.07 0.33 0.34
FamPlnIndex 0.24 0.43 0.46 0.32 0.27 0.33
#Facilities 21 23 4 7 12 67

147

 

Table 4.2 Health Care Characteristics by Final Facility Type

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Public Private Traditional

Distance To Closest (km) 0.66 2.10 0.03
User Fee (1980 pesos) 0.54 4.83 1.06
Doctor Ratio 0.12 0.31 na
Nurse Ratio 0.50 0.54 na

# Personnel 39 49 1
ORT In Stock (0-1) 0.87 0.37 na
Intravenous Diar Trtmts In Stock (0-1) 0.15 0.74 na
Vacc Avl Index (% of 4 basic) 0.44 0.74 na
Farnpln Index (% of 7 methods) 0.38 0.29 na
Educ Medn Tradl In Brgay (yrs) na na 5.09
Frmal ng Medn Tradl Prvdr In Brgay (0-1) na na 0.78

 

148

 

Table 4.3 Utilization by Demographic Group and Season

Type ef Health Eeeility Visited
Public Private Tradl No Visit Obs
All 11.66 17.05 20.46 50.83 31030
Male 11.79 18.14 20.60 49.47 16425
Female 11.51 15.83 20.32 52.33 14541
Low Educ 10.02 11.95 22.99 55.05 9701
High Educ 12.40 19.37 19.31 48.92 21329
Low Asst 13.32 13.04 21.38 52.26 9545
High Asst 10.92 18.83 20.05 50.20 21485
Age 1-6 Mos 18.39 21.61 8.78 51.22 6492
Age 7-24 Mos 9.87 15.84 23.55 50.73 24538
Dry Season 11.67 17.81 20.52 50.00 10407
Wet Season 11.72 16.32 21.11 50.86 12973
Interim Season 11.54 17.25 19.28 51.92 7650

 

Table 4.4 Reported Illness by Demographic Group and Season
WWW
Any Diarrhea FRI Other
All .61 .08 .08 .47
Male .62 .09 .08 .47
Female .61 .07 .07 .48
Low Educ .63 .09 .09 .47
High Educ .61 .08 .07 .47
Low Asst .63 .10 .09 .47
High Asst .60 .08 .07 .47
Age 1-6 Mos .53 .05 .05 .43
Age 7-24 Mos .63 .09 .08 .48
Dry Season .63 .08 .08 .49
Wet Season .60 .08 .07 .47
Interim Season .62 .08 .09 .48

149

Table 4.5

Child Variables
Child's Sex
Child's Age

Household Variables
Mother's Education
Mother's Age

Mother's Height
Husband's Education
Husband's Age

Mother Not in HHold
Husband Not in HHold
Mother Senior Woman
Grandmother Present

# Younger Males

# Younger Females

# Older Males Under 6
# Older Females Under 6
Male Children Age 6-13
Female Children Age 6-13
Males Age 14-20
Females Age 14-20
Males Age 21-60
Females Age 21-60
Males Age 60 or Over
Females Age 60 or Over
Household Assets
Health Insurance

Health Facility Variables
Distance to Public/Private

Distance to Traditional
Facility UserFees

Postnatal Provided
Number Personnel
Doctor Ratio

Nurse Ratio

ORT Availability

Other Diar Trtrnt Avail
Vaccin Availability Index

Exogenous Variables for Analysis of Health Care
Demand—Cebu, Philippines 1983-1986

1=Male, 0=Fema|e
Days Since Birth

Highest Grade Completed

Years

Cm

Highest Grade Completed

Years

Mother never present in hhold during child's ﬁrst 2 years.
Husband never present in hhold during child's ﬁrst 2 years
Mother is Hhold Head, Spouse of Head, or Mother of Head
Sample Woman's Mother or Mother-in-law Present

# Males Younger than Sample

# Females Younger than Sample Child

# Males Older than Sample Child but Ynger than 6

# Females Older than Sample Child but Ynger than 6

# Males Age 6-13

# Females Age 6-13

# Males Age 14-20

# Females Age 14-20

# Males Age 21-59

# Females Age 21-59

# Males Age 60 or Over

# Females Age 60 or Over

Deﬂated value of total hhold asset holdings/1000 in 1980 pesos
Child covered by health insurance

Km from Cluster Location of Household to Nearest Public or
Private Facility

Km to Nearest Traditional Facility from Barangay Center
Barangay Median Deﬂtd I-IH Expenditure Per Visit by Facility
Type in 1980 pesos

Facility provides child postnatal services

Total number personnel at facility

Proportion of doctors to total staff

Proportion of nurses to total staff

Facility has ORT Supplies (0-1)

Facility has intraveneous diarrhea treatments (0- 1)

Percentage of four basic child vaccinations facility has on hand

150

Family Planning Avail Index
Yrs Educ Medn Trad Prvdr

Formal Training Tradl

Percentage of seven possible family planning supplies/treatments
facility provides

Median years of education of traditional providers in the
community

Median training of traditional providers in the community

Non-Facility Community Variables

Piped/Pumped Water

Sanitary Garbage
Water Shortage
Improved Roads
Bank Dummy
High Elevation
Island

Corn Price

Vegetable Oil Price
Minimum Formula Price

Rainfall

% sample hholds in barangay w piped or pumped water into
house or yard

% sample hholds in barangay w sanitary garbage disposal
Frequent water shortages experienced in barangay

Barangay has concrete or asphalt roads

Barangay has modem or "rural" bank

High Elevation Barangay

Island Barangay

Deﬂated Bimonthly Barangay corn prices per kilogram in 1980
pesos

Deﬂated Bimonthly Barangay Vegetable Oil prices per ml in
1980 pesos

Minim Deﬂated Bimonthly Barangay Infant Formula Price per
ml in 1980 pesos

Monthly Levels for Region (mm)

151

 

 

Table 4 . 6 Summary Statistics

 

 

Variable Mean Std. Dev.
1 1. .1 1 n! E I 1‘

Child is Male 0.53 0.50
Child Age in Months 12.73 6.92
Mother Absent All 2 Yrs of Survey 0.00 0.05
Mother Educ Years 7.36 3.71
Mother Age Years 27.43 6.19
Mother at cm 150.73 5.10
Husband.Absent All 2Yrs of Survey 0.04 0.19
Husband Educ Years 7.43 4.10
Husband Age Years 29.18 8.78
Child Covered by Insurance 0.34 0.47
AssetValue-1980 Pesos/1000 8.19 28.36
Mother Is Senior woman in Hhold 0.81 0.39
Grandmother Present in Hhold 0.21 0.41
# Younger Males In HH 0.04 0.21
# Younger Females in HR 0.04 0.20
# Older Under 6 Yr Male 1.06 0.85
# Older Under 6 Yr Feml 0.99 0.84
# Males 6-13 0.56 0.84
# Females 6-13 0.56 0.84
# Males 14-20 0.32 0.67
# Females 14-20 0.43 0.72
# Males 21-60 1.22 0.67
# Females 21-60 1.22 0.74
#iMales Over 60 0.06 0.24
# Females Over 60 0.08 0.27
: '1 --u “E 1.:

High Elevation Brgy 0.06 0.24
Island Barangay 0.10 0.31
Comm Avg Piped/Pumped'Wtr to House 0.13 0.08
Comm Avg Sanitry Garb Disposal Prac 0.79 0.10
Comm Has Freq‘water Shortages 0.14 0.35
Cemm Has Improved Roads 0.73 0.44
Community Has Bank 0.21 0.41
Comm Avg Fridge Owners 0.07 0.04
Comm.Mdn Time To water Source 2.04 2.24
Comm Avg Modern Toilet in House 0.39 0.20
Corn Price-Current Mth 198.51 33.05
Veg Oil Prc-Currenthth 1.02 0.33
Formula Pro-Current.Mth 2.72 0.48
Rainfall mm Current Mth 145.56 99.22

 

number of children: 2884
Number of child-round observations: 30919

152

Table 6.1.

Any Child Curative Health Care Visit:

Individual- and Household-Level Determinants

 

(Full Distance, User Fee, Facility Quality Speciﬁcaton)

 

 

coef z-stat
Child is Male 0.148 3.30
Child Age Mos 1-6 0.142 10.30
Child Age Mos 7-24 -0.040 1 1.22
Mother Absent All 2Yrs -0.435 1.02
Mother Educ 0-5 Years 0.040 1.96
Mother Educ >= 6 Yrs 0.016 2.01
Mother Age (Years) -0.012 2.47
Mother Ht to 150 cm 0.019 2.43
Mother Ht>=150 cm -0.007 1.15
Husband Absent All 2Yrs -0. 126 0.72
Husband Educ 0-5 Years -0.008 0.39
Husband Educ >= 6 Yrs -0.011 1.52
Husband Age (Years) 0.002 0.39
Insurance 0.033 0.82
AssetValue-LowestThird 0.23 1 2.99
AssetValue-UpperTonhrd -0.001 1 .05
Mother Is Senior Woman 0.019 0.25
Grandmother Present 0.033 0.46
# Younger Males In HH -0.214 2.94
# Younger Females in HH -0.113 1.42
# Older Under 6 Yr Male -0.004 0.15
# Older Under 6 Yr Feml 0.035 1.42
# Males 6-13 0.003 0.12
# Females 6-13 -0.015 0.69
# Males 14-20 0.008 0.28
# Females 14-20 0.029 1.16
# Males 21-60 -0.014 0.48
# Females 21-60 0.033 1.19
# Males Over 60 -0. 141 1.81
# Females Over 60 0.021 0.30
Constant Term -5.407 3.09

 

 

Notes: 1. Outcome of no visit is the comparison group. 2. Regression standard errors
adjusted for clustering on individual id. 3. Also included as regressors are current and
lagged rainfall levels for the area, and the following community variables: deﬂated current
and lagged food prices, high elevation indicator; island indicator; proportion of households
in barangay with piped or pumped water into house or yard; proportion of households in
barangay with sanitary garbage disposal; dummy for frequent water shortages; dummy for
presence of improved (paved or concrete) roads; and dummy for presence of formal or rural
banks.

153

Table 6.2.

Any Child Curative Health Care Visit:

Health Provider Attributes, Included in Succesive Steps

 

 

 

Dist, Fee, Quality Distance and Fee Distance Only I
coef z-stat coef z-stat coef z-stat

Dist Nearest Public(km) -0.068 1.33 -0.016 0.40 -0.097 2.68
Price Public Visit 0.543 1.85 0.579 4.16

Number Personnel Public -0.004 3.21

Doctor Ratio Public 0.828 1.64

Nurse Ratio Public 0.171 0.61

ORT Available Public 0.358 2.02

OthrDiarDrug Avl Public 0.363 1.12

Vacc Avail Indx Public 0.448 3.30

Famp Avail Indx Public -0. l 88 0.76

Dist Nearst Private(km) -0.095 3.60 -0.060 3.70 -0.036 2.34
Price Private Visit 0.015 0.27 -0.036 1.69

PM Provides PostNataI -0.095 0.26

Number Personnl Private 0.005 2.97

Doctor Ratio Private 1.403 2.7]

Nurse Ratio Private 1.610 2.88

ORT Available Private -0.215 1.03

OthrDiarDrug Avl Private 0.250 1.37

Vacc Avail lndx Private -0.419 0.82

Famp Avail Indx Private 0.053 0.18

Dist Nearst Traditl(km) -0.732 1.10 -0.308 0.89 -0.722 3.75
Price Traditional Visit 0.074 0.54 0.190 1.93

YrsEduc MedianTrlervdr 0.013 1.22

FormlTraing MedianTrdl -0.182 0.89

Constant Term -5.407 3.09 -2.174 1.73 -1.335 1.08

 

 

 

 

Notes in Table 6.1 apply here.

154

Table 6.3.

Any Child Curative Health Care Visit:

Provider Attributes Included in Succesive Steps & Community Controls

 

 

 

Excluded
Dist, Fee, Quality Dist and Fee Distance Only
coef z-stat coef z-stat coef z-stat
Dist Nearest Public(km) -0.060 1.54 -0.009 0.30 -0.072 2.61
Price Public Visit 0.632 3.06 0.325 3.23
Number Personnel Public 0.000 0.08
Doctor Ratio Public -0.022 0.06
Nurse Ratio Public -0.061 0.38
ORT Available Public 0.338 2.23
OthrDiarDrug Avl Public 0.041 0.28
Vacc Avail Indx Public 0.069 0.81
Famp Avail Indx Public -0.295 1.47
Dist Nearst Private(km) -0.119 7.85 -0.081 7.21 -0.077 7.57
Price Private Visit -0.059 2.27 -0.015 1.09
Prvt Provides PostNatal -0.491 3.65
Number Personnl Private 0.004 2.82
Doctor Ratio Private 0.622 1.55
Nurse Ratio Private 1.054 2.49
ORT Available Private -0.188 1.00
OthrDiarDrug Avl Private 0.1 12 0.87
Vacc Avail Indx Private -0.255 0.61
Famp Avail Indx Private -0.609 2.30
Dist Nearst Traditl(km) -0.127 0.34 -0.230 1.24 -0.338 2.91
Price Traditional Visit 0.274 3.01 0.224 3.78
YrsEduc MedianTrlervdr 0.005 0.71
FormlTraing MedianTrdl 0.119 0.79
Constant Term -4. 168 -3.44 -3 .439 -2.94 -2.998 -2.58

 

 

 

 

Notes: 1. Outcome of no visit is the comparison group. 2. Regression standard errors
adjusted for clustering on individual id.

155

Table 6.4. Baseline Specification of Facility Choice for Child Curative Visit

 

Multinomial Logit Regression

 

 

Public y None Erivate y N911: [raditiggal y None
Variable Coef . z Coef. z Coef. 2
Public facility Characteristics
Nearest Public (km) -0.360 4.46
Price Public Visit 0.092 0.31
Number Personnel Public 0.006 4.55
Doctor Ratio Public 1.544 3.41
Nurse Ratio Public 0.116 0.38
ORT Avail Indx Public 1.042 4.72
OthrDiarDrugIndx Public -1.221 4.16
Vacc Avail Indx Public 0.615 4.37
Famp Avail Indx Public 0.335 1.13
Nearest Private (km) . -0.069 2.42
Price Private Visit -0.017 0.81
Prvt Provides PostNatal 0.439 1.1 1
Number Personnl Private 0.002 1.25
Doctor Ratio Private -0.400 1.05
Nurse Ratio Private -0.230 0.52
ORT Avail Indx Private -0.403 2.00
OthrDiarDruglndx Privat 0.305 2.12
Vacc Avail Indx Private -0.757 1.92
Famp Avail Indx Private -0.601 2.09
Nearest Traditiona1(km) -1.990 5.41
Price Traditional Visit 0.241 2.10
YrsEduc MedianTrlervdr 0.010 1.31
FormlTraing MedianTrdl 0.466 6.42
Child is Male 0.098 1.34 0.200 3.08 0.122 2.20
Child Age Mos 1-6 0.042 2.15 0.121 6.72 0.372 15.15
Child Age Mos 7-24 -0.082 12.66 -0.063 12.52 -0.002 0.33
Mother Absent A11 2Yrs -0.227 0.43 -0.206 0.36 -0.924 1.57
Mother Educ 0-5 Years 0.039 1.17 0.083 2.41 0.039 1.55
Mother Ecuc >= 6 Yrs 0.020 1.48 0.028 2.70 -0.002 0.20
Mother Age (Years) -0.021 2.51 -0.009 1.28 -0.009 1.42
Mother Ht to 150 cm 0.023 1.82 0.030 2.71 0.004 0.46
Mother Ht>=150 cm -0.021 2.04 -0.004 0.46 -0.004 0.62
Husband Absent All 2Yrs 0.194 0.68 0.251 0.99 -0.575 2.55
Husband Educ 0-5 Years 0.054 1.56 0.006 0.17 -0.032 1.35
Husband Educ >= 6 Yrs -0.029 2.22 0.012 1.23 -0.025 2.74
Husband Age (Years) 0.008 1.19 0.005 0.69 -0.005 1.0

156

Table 6.4 (cont’d)

 

 

Variable Coef . z Coef. z Coef. 2
Insurance -0.087 1.27 0.192 3.42 -0.016 0.30
AssetValue—LowestThird 0.018 0.14 0.667 5.89 0.034 0.36
AssetValue-UpperTonhrd —0.005 2.38 0.001 1.04 -0.002 1.99
Mother ls Senior Woman -0.121 0.96 0.119 1.15 -0.060 0.65
Grandmother Present 0.045 0.37 0.067 0.69 -0.019 0.20
# Younger Males In HH -0.063 0.47 -0.365 3.04 -0.195 2.32
# Younger Females in HH 0.092 0.62 -0.184 1.44 -0.120 1.29
# Older Under 6 Yr Male 0.026 0.66 -0.045 1.28 0.029 0.95
# Older Under 6 Yr Feml 0.066 1.66 -0.012 0.34 0.067 2.16
# Males 6-13 0.043 1.13 -0.019 0.57 0.006 0.20
# Females 6-13 0.011 0.29 -0.014 0.44 -0.020 0.70
# Males 14-20 0.022 0.50 -0.034 0.94 0.036 1.03
# Females 14-20 -0.078 1.74 0.091 2.76 0.019 0.59
# Males 21-60 -0.012 0.24 0.019 0.47 -0.046 1.29
# Females 21-60 0.012 0.26 0.071 2.02 0.000 0.01
# Males Over 60 -0.077 0.61 -0.189 1.82 -0.168 1.66
# Females Over 60 0.018 0.14 0.036 0.38 0.044 0.50
Constant Term -4.514 2.15 -5.482 2.93 -2.836 1.82
Chi-Square Test Statistics for Joint Signiﬁcance of Variable Groups
Facility 132.49 0.00 23.97 0.00 58.37 0.00
Fee and Distance 21.17 0.00 6.04 0.05 29.86 0.00
Personnel 47.76 0.00 2.61 0.46 44.72 0.00
Drug Availability 107.79 0.00 11.02 0.03 na na
Individual and Household
Child Age 164.62 0.00 164.04 0.00 256.16 0.00
Mother Educ 4.75 0.09 18.18 0.00 2.47 0.29
Mother Height 5.30 0.07 7.98 0.02 0.43 0.81
Husband Educ 5.80 0.05 1.74 0.42 12.45 0.00
Value Assets 5.68 0.06 36.26 0.00 3.97 0.14
Hhold Composition 13.38 0.50 31.36 0.00 17.01 0.26
Community 146.28 0.00 124.55 0.00 193.27 0.00
Infrastructure 93.94 0.00 26.20 0.00 83.75 0.00
Food Prices 10.05 0.35 21.78 0.01 39.32 0.00
Rainfall 15.91 0.26 61.14 0.00 39.75 0.00
Number of obs = 30919
Pseudo R2 = 0.0636

Log Likelihood = -35349.89

 

Notes in Table 6.1 apply here.

157

Table 6.5.

Facility Choice for Child Curative Care: Individual and Household-Level

Determinants with Barangay-Year Interaction Control Dummies

 

 

 

A11 Hhold- & Insurance Hhold Comp Hhold Comp &
Indiv-Level Excluded Excluded Insur Excluded
coef z coef coef z _ coef 2

Public Visit vs No Visit
Child is Male 0.080 1.09 0.081 1.10 0.047 0.80 0.048 0.81
Child Age Mos 1-6 0.00] 2.20 0.001 2.21 0.001 2.12 0.001 2.13
Child Age Mos 7-24 -0.003 11.77 -0.003 11.77 -0.003 12.26 -0.003 12.26
Mother Absent All 2 Yrs -0.158 0.30 -0.189 0.37 0.036 0.07 0.010 0.02
Mother Educ 0-5 Years 0.044 1.29 0.044 1.27 0.045 1.31 0.044 1.29
Mother Educ >= 6 Yrs 0.021 1.58 0.020 1.48 0.018 1.36 0.016 1.24
Mother Age (Years) -0.021 2.46 -0.021 2.48 -0.018 2.42 -0.018 2.44
Mother Ht to 150 cm 0.025 1.95 0.024 1.93 0.025 2.00 0.025 1.97
Mother Ht>=150 cm -0.020 1.91 -0.020 1.92 -0.020 1.93 -0.020 1.95
Husband Absent A11 2Yrs 0.134 0.47 0.116 0.40 0.188 0.66 0.168 0.59
Husband Educ 0-5 Years 0.035 1.05 0.034 1.02 0.036 1.08 0.035 1.04
Husband Educ >= 6 Yrs -0.022 -1.71 -0.024 1.92 -0.023 1.80 -0.025 2.05
Husband Age (Years) 0.008 1.21 0.008 1.15 0.008 1.20 0.008 1.13
AssetValue-LowestThird 0.018 0.14 0.008 0.06 0.059 0.47 0.048 0.39
AssetValue-UpperTonhrd -0.005 2.19 -0.005 2.16 -0.004 2.19 -0.004 2.16
Insurance -0.062 0.89 -0.073 1.04
Mother Is Senior Woman -0.074 0.60 -0.080 0.64
Grandmother Present 0.075 0.63 0.073 0.62
# Younger Males In HH -0.020 0.15 -0.017 0.12
# Younger Females in HH 0.063 0.43 0.062 0.42
# Older Under 6 Yr Male 0.029 0.73 0.030 0.75
# Older Under 6 Yr Feml 0.067 1.70 0.068 1.70
# Males 6-13 0.044 1.16 0.046 1.20
# Females 6-13 -0.003 0.07 -0.001 0.03
# Males 14-20 0.031 0.70 0.032 0.73
# Females 14-20 -0.063 1.43 -0.064 1.46
# Males 21-60 -0.014 0.29 -0.013 0.28
# Females 21-60 0.003 0.07 0.002 0.04
# Males Over 60 -0.055 0.43 -0.057 0.45
# Females Over 60 -0.026 0.20 -0.023 0.18
Constant Term -4.816 2.50 -4.745 2.47 -4.913 2.55 -4.841 2.53

 

 

 

 

 

158

 

Table 6.5 (continued)

 

 

 

All Hhold- & Insurance Hhold Comp Hhold Comp &
Indiv-Level Excluded Excluded Insur Excluded
coef coef z coef z coef

Private vs. No Visit
Child is Male 0.217 3.33 0.214 3.28 0.170 3.48 0.168 3.44
Child Age Mos 1-6 0.003 6.19 0.003 6.18 0.003 6.44 0.003 6.44
Child Age Mos 7-24 -0.003 13.43 -0.002 3.40 -0.003 14.26 -0.003 14.24
Mother Absent A11 2 Yrs -0.028 0.05 0.053 0.10 -0.122 0.21 -0.053 0.10
Mother Educ 0-5 Years 0.087 2.47 0.089 2.54 0.081 2.32 0.084 2.40
Mother Educ >= 6 Yrs 0.028 2.68 0.032 3.05 0.033 3.12 0.038 3.54
Mother Age (Years) -0.010 1.36 -0.009 1.29 -0.01 1 1.53 -0.010 1.44
Mother Ht to 150 cm 0.030 2.71 0.030 2.75 0.031 2.81 0.031 2.85
Mother Ht>=150 cm -0.003 0.36 -0.002 0.23 -0.003 0.37 -0.002 0.23
Husband Absent A11 2Yrs 0.251 1.00 0.316 1.26 0.283 1.13 0.341 1.37
Husband Educ 0-5 Years 0.004 0.11 0.008 0.22 0.004 0.11 0.008 0.24
Husband Educ >= 6 Yrs 0.011 1.07 0.017 1.76 0.012 1.20 0.019 1.94
Husband Age (Years) 0.005 0.80 0.006 0.98 0.005 0.80 0.006 0.96
AssetValue-LowestThird 0.678 5.95 0.709 6.23 0.699 6.22 0.723 6.45
AssetValue-UpperTonhrd 0.001 1.36 0.001 1.28 0.001 1.90 0.001 1.79
Insurance 0.198 3 .50 0.207 3 .69
Mother Is Senior Woman 0.128 1.23 0.147 1.42
Grandmother Present 0.076 0.77 0.079 0.80
# Younger Males In I-IH -0.383 3.23 -0.397 3.36
# Younger Females in HH -0.201 1.56 -0.205 1.60
# Older Under 6 Yr Male -0.051 1.43 -0.051 1.42
# Older Under 6 Yr Feml -0.002 0.06 -0.004 0.1 1
# Males 6-13 -0.018 0.53 -0.022 0.67
# Females 6-13 -0.009 0.29 -0.013 0.42
# Males 14-20 —0.046 1.25 -0.051 1.39
# Females 14-20 0.101 3.06 0.104 3.15
# Males 21-60 0.014 0.35 0.012 0.30
# Females 21-60 0.074 2.12 0.079 2.27
# Males Over 60 -0.166 1.61 -0.156 1.54
# Females Over 60 0.042 0.44 0.034 0.35
Constant Term -6.798 4.10 -6.929 4.17 -6.753 4.09 -6.883 -4.16

 

 

 

 

 

159

 

Table 6.5 (continued)

 

 

 

 

 

 

 

 

 

 

All Hhold- & Insurance Hhold Comp Hhold Comp &
Indiv-Level Excluded Excluded Insur Excluded
coef coef z coef z coef 2
Traditional vs No Visit
Child is Male 0.124 2.24 0.124 2.25 0.088 2.04 0.089 2.05
Child Age Mos 1-6 0.014 19.70 0.014 19.70 0.014 19.78 0.014 19.78
Child Age Mos 7-24 0.000 2.17 0.000 2.17 0.000 3.07 0.000 3.07
Mother Absent A11 2 Yrs -0.914 1.50 -0.916 1.50 -0.831 1.38 -0.832 1.38
Mother Educ 0-5 Years 0.025 0.99 0.024 0.99 0.021 0.86 0.021 0.86
Mother Educ >= 6 Yrs -0.006 0.66 -0.007 0.68 -0.008 0.81 -0.008 0.83
Mother Age (Years) «0009 1.43 -0.009 1.43 -0.007 1.27 -0.007 1.27
Mother Ht to 150 cm 0.007 0.76 0.007 0.76 0.008 0.83 0.008 0.83
Mother Ht>=150 cm -0.003 0.40 -0.003 0.41 -0.003 0.41 -0.003 0.42
Husband Absent All 2Yrs -0.657 2.93 -0.658 2.94 -0.633 2.88 -0.634 2.88
Husband Educ 0-5 Years -0.037 1.52 -0.037 1.52 -0.035 1.45 -0.035 1.46
Husband Educ >= 6 Yrs -0.024 2.61 -0.024 2.68 -0.025 2.68 -0.025 2.75
Husband Age (Years) -0.007 1.36 -0.007 1.37 -0.007 1.41 -0.007 1.42
AssetValue-LowestThird 0.123 1.29 0.122 1.28 0.127 1.36 0.126 1.35
AssetValue-UpperTonhrd -0.002 1 .90 -0.002 1 .88 -0.002 2.04 -0.002 2.02
Insurance -0.007 0.13 -0.006 0.1 1
Mother Is Senior Woman -0.064 0.68 -0.064 0.68
Grandmother Present -0.041 0.45 -0.041 0.45
# Younger Males In I-IH -0.206 2.48 -0.206 2.48
# Younger Females in HH -0.152 1.65 -0.152 1.66
# Older Under 6 Yr Male 0.018 0.56 0.018 0.56
# Older Under 6 Yr Feml 0.057 1.84 0.057 1.84
# Males 6-13 0.005 0.16 0.005 0.16
# Females 6-13 -0.023 0.81 -0.022 0.81
# Males 14-20 0.034 0.95 0.034 0.95
# Females 14-20 0.013 0.40 0.013 0.40
# Males 21-60 -0.046 1.29 -0.046 1.29
# Females 21-60 0.002 0.05 0.002 0.05
# Males Over 60 -0.154 1.53 -0.154 1.52
# Females Over 60 0.039 0.44 0.040 0.45
Constant Term -3.773 2.65 -3.774 2.66 -3.912 2.76 -3.914 -2.76
LR Test vs. Full Set Indiv and HI-Iold Vars Stat Prob Stat Prob Stat Prob
111.31 0.00 36.46 0.00 153.03 0.00

 

 

 

 

Notes: 1. Outcome of no visit is the comparison group. 2. Standard errors adjusted for

clustering on id.

160

 

Table 6.6. Facility Choice for Child Curative Care Visit: Provider Attributes Included in
_ Succesive Steps w Full Se_t of Community Controls

 

 

 

 

 

 

 

 

 

___ ___ a +2ist, Fee, Quality Dist and Fee Distance Only _
___:991 _ _ ¥-_s_t§t_________coef ertatj . “.6221“ _z-§tat
Public Visit vs No Visit
Dist Nearest Public (km) -0.360 4.46 -0.352 4.51 -0.363 4.71
Price Public Visit ' 0.092 0.31 0.130 0.70
Number Personnel Public 0.006 4.55
Doctor Ratio Public . 1.544 3.41
Nurse Ratio Public 0.116 0.38
ORT Available Public 1 1.042 4.72
OthrDiarDrug Avl Public -1.221 4.16
Vacc Avail Indx Public 0.615 4.37
F amp Avail Indx Public 0.335 1.14
Constant Term -4.514 2.15 . -3.222 1.62 -3.159 1.59
Private Visit vs No Visit
Dist Nearst Private (km) -0.069 2.43 -0.102 4.29 -0.099 4.22
Price Private Visit -0.017 0.82 -0.020 1.23
Prvt Provides PostNatal 0.439 1.1 1
Number Personnl Private 0.002 1.25
Doctor Ratio Private ' -0.400 1.05
Nurse Ratio Private -0.230 0.52
ORT Available Private -0.403 2.00
OthrDiarDrug Avl Privat 0.305 2.13
Vacc Avail Indx Private -0.757 1.92
Famp Avail Indx Private -0.601 2.09
Constant Term -5.482 2.93 -5.073 2.87 -5.358 3.05
Traditional Visit vs No Visit
Dist Nearst Traditl (km) -1.990 5.41 -1.369 4.59 -1.298 4.67
Price Traditional Visit . 0.241 2.10 0.035 0.32
YrsEduc MedianTrlervdr 0.010 1.31
FormlTraing MedianTrdl 0.466 6.42
Consent Term 5. . __-2_-8_3§_ - -1;82.__;2_-16_1_ _ _1-7_§l_ 225____ _1.75__

 

Notes in Table 6.1 apply here.

161

 

Table 6.7.1

Facility Choice for Child Curative Care Visit:
Provider Attributes Included in Succesive Steps & Community Controls

 

 

 

 

 

 

 

 

Excluded
Dist, Fee, Quality Dist and Fee Distance Only

_ __ _ coef z-stat coef z-stat coef z—stat
Public Visit vs No Visit
Dist Nearest Public (km) -0.465 8.28 -0.522 9.26 -0.527 9.03
Price Public Visit -0.386 1.81 -0.888 6.28
Number Personnel Public 0.006 7.16
Doctor Ratio Public 1.238 3.20
Nurse Ratio Public 0.248 1.55
ORT Available Public 0.989 5.27
OthrDiarDrug Avl Public -1.069 6.62
Vacc Avail Indx Public 0.146 1.32
Famp Avail Indx Public 1.026 4.25
Constant Term -6.256 3.28 ~3.531 1.88 -4.235 2.23
Private Visit vs No Visit
Dist Nearst Private (km) -0.134 7.48 -0.130 8.40 -0.125 8.44
Price Private Visit 0.006 0.35 0.008 0.72
Prvt Provides PostNatal 0.056 0.40
Number Personnl Private 0.001 0.60
Doctor Ratio Private -0.722 3.07
Nurse Ratio Private -0.683 2.18
ORT Available Private -0.162 0.84
OthrDiarDrug Avl Private 0.031 0.32
Vacc Avail Indx Private -0.360 1.00
F amp Avail Indx Private -0.460 1.74
Constant Term -5.788 3.44 -6.395 3.83 -6.3 78 3.81
Traditional Visit vs No Visit
Dist Nearst Traditl (km) -0.248 1.54 -0.045 0.29 -0.085 0.55
Price Traditional Visit 0.029 0.45 -0.018 0.31
YrsEduc MedianTrlervdr 0.009 1.63
F ormlTraing MedianTrdl 0.245 4.30 *
Constant Term -4.066 2.83 -3.923 2.75j -3.947 2.78

 

 

Notes: 1. Outcome of no visit is the comparison group. 2. Regression standard errors

adjusted for clustering on individual id.

162

 

Table

6.7 .2 Facility Choice for Child Curative Care Visit: Provider Attributes Included in

 

 

__ Succesive Steps w Community Controls Replaced by Municipality Dummies_

 

 

 

___ Dist, Fee, Quality Dist and Fee Distance Only _
____ coef z-stat coef z-stat: coef z-s_tig
Public Visit vs No Visit
Dist Nearest Public (km) -0.627 -8.08 ~0.504 -7.81 -0.483 -7.43
Price Public Visit -0.432 -1 .44 -0.789 -5.17
NumberPersonnel Public 0.005 3 .70
Doctor Ratio Public 1.885 3.01
Nurse Ratio Public 1.006 4.33
ORT Available Public 1.175 4.13
OthrDiarDrug Avl Public -0.646 -2.51
Vacc Avail Indx Public 0.234 1.84
Famp Avail Indx Public 0.878 3.13
Constant Term -6.3 76 -3.34 -3.684 -1.97 -4.179 -2.22
Private Visit vs No Visit ‘
Dist Nearst Private (km) -0.128 -6.88 -0.115 -6.96 -0.110 -6.95
Price Private Visit -0.006 -0.32 0.004 0.29
Prvt Provides PostNatal 1.487 3.92
Number Personnl Private 0.000 0.06
Doctor Ratio Private -0.963 -2.37
Nurse Ratio Private -0.897 -2.17
ORT Available Private -0.063 —0.31
OthrDiarDrug Avl Privat -0.329 -1.46
Vacc Avail Indx Private 0.104 0.20
Famp Avail Indx Private -0.500 -1 .64
Constant Term -6.886 -4.02 -6.207 -3.72 -6.178 3.71
Traditional Visit vs No Visit
Dist Nearst Traditl (km) -0.237 -1.32 -0. 173 -0.99
Price Traditional Visit 0.009 0.12 -0.041 -0.55
YrsEduc MedianTrlervdr -0.005 -0.82
Fonanraing MedianTrdl 0.061 0.96
Constant Term -3.888 -2.72 -3.808 -2.68 -0.182 -1.06

 

 

 

 

Notes: 1. Outcome of no visit is the comparison group.
for clustering on individual id.

163

2. Regression standard errors adjusted

 

Table 6.8 Effect of Facility Price and Quality on Choice-
Facility Effects Constrained to Equality

 

Conditional Logit Regression

 

 

variable Coef. 2
Distance -0.167 -l4.76
User Fee -0.026 -2.51
Postnatal Services -0.021 -0.31
Number Personnel 0.000 0.98
Doctor Ratio 0.498 5.34
Nurse Ratio 0.537 6.05
ORT Available -0.190 -5.23
Othr Diar Avail 0.005 0.11
vaccn Avail Indx 0.525 10.63
Fampl Avail Indx -0.339 -4.69
Yrs Educ Tradl 0.001 0.53
Formal Training Tradl 0.235 5.95
Public Intercept -5.043 -4.13
Private Intercept -7.378 -6.49
Traditl Intercept -3.912 -4.07
Number of obs = 30920

Log Likelihood = -35938.261

Pseudo R2 = 0.1616

 

Notes: 1. Standard errors NOT adjusted for clustering on id.
2. Results of household effects not reported in this table
because they are nearly identical to those in "unconstrained"
specification.

164

 

165

 

 

mmo.o Hao.ol ooo.o ouo.ol moo.ot eta up: one eaaeo

Hao.o oHo.o «mo.o ooo.o moo.o can: an passe

Huao Mboo mmoou goon hvo.0l moo.o.. 3&5;ng ocwnquﬁ—om
Hoo.0l uoo.o ooo.o ooo.o ooo.o ooo.o uppumauuaeaaunz newsman
HHo.o mmo.o noo.ot soo.ou voo.ot moo.ol unna> Hucowuaunua oowum
oam.ot oam.ol omo.o «mo.o mHH.o mmo.o laxvauauuua unanoz page
«no.0 mac.o ouo.ot Hmo.ot moo.o oao.o ouu>wum even dunks mans
mmo.o vuo.o ooo.o- ~oa.ou mHo.o «Ho.o sunbeam area Hau>4 oou>
umo.ol aoo.ot mmo.o avo.o ooo.o: moo.ol unvaum and usuouawouruo
ouo.o mao.o evo.ou wmo.ol oHo.o ooo.o ouubaum manaawuaa amo
poo.o poo.o HHo.ot Hmo.Ol moo.o voo.o anabaum oases onusz
ouo.o «Ho.o mmo.ot vmo.ot poo.o ooo.o «cabana owned uouooo
ooo.o ooo.o ooo.o ooo.o ooo.o ooo.o oun>uum Heeonuom nonapz
ooo.o: vac.ol ooo.o amo.o mao.ot >oo.ol aauuzunom mova>oum abum
ooo.o Hoo.o Hoo.o ~oo.ot ooo.o ooo.o cana> sunbeam oowum
Hoo.o moo.o Hoo.ot moo.ot ooo.o Hoo.o laxvoua>aum unuuoz cane
muo.ol ooo.o- noo.ot moo.ot avo.o omo.o chased seen Hanan gash
muo.ol «no.0- ooo.o: oao.ou bvo.o mmo.o ceased anew Hau>4 ouc>
«Ha.o vmo.o ~mo.o ouo.o mpa.ot oaa.ot ceased and usunuuaouruo
vmo.ol ouo.ot mao.ot pao.ol moa.o «mo.o nausea canaaanpa amo
auo.o «oo.ol ooo.o uoo.on ovo.on oHo.o ceased chasm onusz
upo.ot omo.ot owo.ol muo.ot mHH.o mma.o change canon uouooa
ooo.o ooo.o ooo.o ooo.o Hoo.o Hoo.o chased nonconuom uonapz
mao.ot ~oo.ot voo.ot Hoo.ol muo.o ooo.o usua> chased ooaum
mao.o poo.o voo.o ooo.o muo.ol mmo.ol laxcoaansm unauaaz undo
wv . mun mmm . "moon "cowHUUAMAOOAm vounoz How 09Hd> abdnnHOGH

Heaz an: Haﬁz Hex” Haﬁz an:

202 > EOHBHEB 202 > HagHMm H202 b UHHmDm Hd>

mango: BHUOA §2HB§ OHM—“mg m> §2HH§

”A; QEOOﬁIImBUEH HBHAHU‘E gHUg QOHBHEOUZD m.w 32H.

 

ooo.o ooo.o ooo.o Hoo.o ooo o oo uo>o «causes *
oHo.ot ooo.o- oHo.ol Hoo.ot Hoo.ot oo uo>o «can: u
ooo.o- ooo.o ooo.o ooo.o ooo.o ootaa «cannon *
ooo.o- ooo.o ooo.o ooo.o ooo.o ootau «can: *
ooo.o ooo.o mHo.o ooo.o- ooo.o- omtoﬁ nuances *
ooo.o ooo.o- ooo.o- uoo.ot ~oo.o outoa «can: a
ooo.o- ooo.o aoo.ol ~oo.o ~oo.o mate «nausea a
Hoo.o ~oo.ol ooo.o- ooo.o ooo.o male noun: *
ooo.o ooo.o- ooo.o- ooo.o moo.o Have a» o sauce noose a
moo.o ooo.o- ooo.o- ooo.o ooo.o can: u» o noose nooao *
mao.ou oao.ou muo.ol «Ho.o oao.o mm on «cameos nausea» o
oao.ot ouo.ot ~vo.ol ooo.o: ooo.o us ca «can: uoucsou *
ooo.o- ooo.o ooo.o ooo.o ooo.o ocunaum noruoaeeuuo
Hao.ol «Ho.o ooo.o «Ho.ot «Ho.ot eases uoacam an aoruoz
ooo.o ooo.o ooo o ooo.o ooo.o unraoxauoaaotonau>uonu4
oao.ot ooo.o ooo.o ooo.o oHo.ot ouanauuoxoqtosau>uonna
ooo.o- ouo.o ooo.o ooo.o- ﬁao.ol ooeuusuca
Hoo.ot Hoo.o Hoo.o Hoo.o Hoo.o .nuuo». one assumes
ooo.o- ~oo.o ooo.o Hoo.ot ~oo.ou an» o nA uses cannons
ooo.o- ooo.o Hoo.o ooo.o ooo.o «noon muo anon oneness
HoH.ot Hoo.o ooo.o moo.o ooo.o mama Ham unann4_oeanunm
ooo.o ooo.o ooo.o ooo.o- ~oo.ot so omHuAum wagons
Hoo.ot ooo.o ooo.o ooo.o ooo.o so one on an Harbor
Hoo.ot Hoo.ot Hoo.ot ~oo.ot aoo.ot Anemone one corner
~oo.o- ooo.o ooo.o ooo.o Hoo.o an» o uA onus earner
ooo.o ooo.o ooo.o ooo.o Hoo.o «new» mto uses morons
mma.ot ooo.o ooo.o Hoo.o aoo.o uuuu Has nuance corner
ooo.o ooo.o- ooo.o- oHo.ou ooo.o- vans no: oma_oaaro
Heaz Hex” Heaz «can Haﬁz Hex

N202 > gOHBHn—gn.

202 > NB¢> HMm

N202 > UHAmDm

Hd>

resonances a .o 3mg

166

 

 

 

 

 

 

 

N: 35 SW 4.2 :8 we we 32 3: 2 98 2: 988833.1338223205
3:. we. as. we EN EN mom as 3: q: q: Nd €5.3313880203205
34 ca 4.8 mi SN mom 5: oi q: a: 8: as am: 0850582825$383M
on. Sm ea. a: mom 3: m2 we we 2m a: mom assassinaaeoaizo
33 a? o9. we Em com o: «.2 92 a: a: ”.2 sonaﬁﬁséaapamowﬁfsa
24 Now as. 2: com 52 3: 92 4.2 n: of 1.2 grassesﬁéegiwﬁfaaoﬁz
SH 3% 3a 2m 2m SN mom ”.2 3: on as 3 2231352323on
2:. Em 92. 92 EN SN 2: we 92 w: 3: 52 23313323220035:
mom Nd mom EN EN EN 3: 1.2 E 2: 2: 5.: 33388.1oeaoaooéoam
mom 2m mom EN EN 2m 3: q: a: 2: we 5: 832532
as: 33 =< ﬁrm 33 =< as: 33 3.. norm 33 3..
55 oz izoEoéo mo<>aa 2.52
.33 .32 238.3: 3 83.6 base a. 825.3%: 33:55 :82 as 2.3

167

 

 

 

So So woo So So So So . .3582 wee-cocoa
”no So :2 So So So So . stag-EB: 03mm;
2 o So So So So So So . ES .2863. 8F.
S? ovo- loo «.5- ooo So So . AEVESEBP aoaoz
huﬁhmsuuuagﬁb LMEASNAN Bﬁgghk
So So So 8. _- omo ooo- ooo . a§E SE .3}. ops-a
So So moo So- So So- So . 225 so... =92 02;
So So So omo So So So . sea Seasoning
So So So who- So 2 o- ooo . 2&5 oz: :92 55
So So So 8o ”2 8o. ooo . 225 25 0.92
So So ooo «_o a: So. So . 29$ coax .989
So So So So :o So So . 295 .285; 8:52
So So So So. So So So . _anzaoo 828$ Em
So So So So woo So- So . ES 39,5 3:;
So go or So- ooo So- So . as: 395 852
humuﬁhmxuﬁhaﬂb awmﬁuﬁtﬁ 39!»:
So So So So So So So So 23.5 oz; =§< 9.5
So So So So So So ooo So 2.95 x2: :92 8s>
So So So So So :._- 43 oo. _- £33 goo—waning
So So So So So So So So 253 x2: .32 b5
So So So So So So So So- 258 cone 2:2
So So So So So 34 ”mo of 23.5 82 cacao
So So So So So So 2o So 23E 358$ .8822
So So So So So ooo So So- 55 235 85
So So So So So So- So So- es: 23.5 .352
assesses-o Smock Sosa
N @000 N HOOD N H000 N H000 N “000 N HOOD
0350..— 032 2mth 2w: o-anm 2m:
so; 325 23.5

 

 

 

note—EU 2:80.,— .o.> 2:: tom 0.30 5.32 95350 230 .3.— ogoau .0523 no 8558.880 5. «Zah-

168

 

 

 

N_d Nod- and mod _ _d Ndd wad Nod mm; VNd Nd; 3d- do 50 5.95-.— an
odd mod- mm; MNd- Nwd 3d- 84 NNd- mod 5.? vwd m _ d- do 5>O .552 a
de Ndd- wmd Ndd NNN Ed and mod 3d 5d vod ddd dc-_N 55:5“— a
mdd cod and 8d- 5d vod NNd 5d 5d mdd- de Nod SAN 532 a
dad mod :d odd 3V.— N-dd 3N add cm; dod- ~N._ 3d- dN-E moo—«Eva a
3..— ddd 2d 5.? mod mdd- bod mod- 2: odd emd Nod- dN-E 532 a
mud med 3..— Ndd- odd ddd Sud 5d- 3d ddd mod mod 3% 5580a a
ddd ddd mod cod 3; add- odd mod dwd mdd- EN N_d 2-0 532 m.
N_.N 3d 2: vod NNd 3.? Ed 3d- de _dd :N N_d 95m ._> c .595 .590 u
add odd 8.. vod m _._ odd- Nnd mod- mod mod- N_._ dd 5.22 ._> o .595 .520 a
dmd mod- _N._ 2d- :N end- de mod- 3..— mmd wed 2d- 2: 5 55:5...“ 5wc=o> n
3; VNd- No; 2d- RN Ed- 34 cNd- omd 2d. 3d Ndd- IE E 55.2 cow—50> a
dN._ Cd- and 3d :d _ _d- GA ~Nd nod _ _d- 3d 2d 955.5 55089.80
do; oNd- ed; 2 d _md add- on; oNd do; VNd- VNd vod- coEo? cot—om mm 5532
Ed 3d- Dod ddd :d odd 2 _._ ddd mmN _dd- m _ ._ ddd oEHoBPaoddDoEo>5mm<
de vod- end d _ d add add m m6 cod mud n _ d m md odd- oumﬁomo>5452o>5mm<
mvd mod.- Sd ddd NcN NNd mmN w _ d t..— w a d- h _ d Ndd- 355m:—
mod 8d- dud _dd- _md ddd nod _dd mod _dd dad _dd $.55er ow< 953::
md._ Nod- coo” mod- NN._ Nod 3nd Ed dmd _dd- MON mod- mo> e HA 26m 953::
34 dd. dvd Nod- do; mdd de mod- 3.. odd med mod 35> mtd 26m 953::
.3.— dmd- Nd.N vod- 34 Nvd de nod wed de 3d 8d mo>N =< “combos 953:2
Md; 8.? odd ddd mud _dd- Ed odd mvN vod- omd 8d- :5 meAI 5:82
N. _ d odd mvd 5d mm; mdd d_.N mod mad Ndd dN._ Nod :5 on _ 8 «3 55oz
we; _dd- end 5.? we; Ndd- 3d ad. ad; ad. ddN Nod- Amuoo>v ow< 5502
3nd Ed Z-d _dd- mN.N mod 34 mod Nd; Ndd we; Nod m; 0 HA 26m 5:82
mN.N add and 8d NN._ odd :.N add mNA bod odd Med 85> mtd comm .5522
dwd odd MNd ddd 36 mod- v93 3d- Nb.» add- vmd mod- VNN moi ow< EEO
ﬁd— wmd v3: de nmé 2d dd;V N_d Nd.N odd do; mod w; mo: ow< EEO

aﬁeﬂsueoesoeb 323.332 93 3:93.55

55th 2oz 295m 032 295m 052
.92..- 8o>tm £32m

 

 

 

€2,569 3 2.:

169

.55: >35 5o 25d. :0 5502

 

 

 

 

 

adthmoT 2.0.3wa 5005—50—5 mo:
mmodd whodd NM 095mm
3 m3 EEO mdvf ”whom 30 .«0 55::2
dod aﬁmm w _ d 3.: Odd _m.\-m ddd mimm wmd 3w: oNd 8.2 :5c5m
_od :umN Odd _m.VN :d _0.N_ cNd 5.: mad 3:“ hmd 3.0 5&5 500m
cod ondm cod 3.? de em. 2 3d mméN ddd omdm cod 2“,: 8:55:55:—
ood mNdZ dod mvddﬁ ddd Nwdw cod ”0.: Odd $.Nw dod mmNo 3.55.5509
odd oo._N w _ d 3.2 mod oo©N Ed w 2 dN cmd 5.2 de 8.3 55500500 20::
cod 3.: odd mud ood ood— dod mde vod dmé ovd om; m5mm< u55>
Ed mod 5d dad ONd mNd mwd mmd VNd wwN vod 2‘6 0:9.”— 955::
nmd 2 ._ mwd de _Nd mod 8d mN.m mod Nod vvd we; Emmi 5502
mod Nod 2d mmd _dd mvd dod dod— Ed omxc. mmd N_.N gum .5502
Odd om.mN_ ddd 3N2 cod no. _ w ddd 9H: ddd den ood moNo ow< EEO
30.3.30: av 339955
5: 5: 5c a: 3d w 2 .2 and :4 Odd Nwdv ood moNc b:55_5>< MEG
cod won cod vo.m_ NNd NOV vmd wm.m ddd Q-dN Odd SdN 5::0m5m
cod emdN 5d 3.3 amd a; mod mos cod 3.: Nod No.5 8555 95 5m
odd 3.: Odd cN. _N 5d ENN mod >3: ood .5960 Odd Em: 55.03%
3:050 533...; H0 wonwognﬂw «50H :5 553me 55,—. aha—_vw-EU
no; dmN- 32 3d- M3.— wmw- _NN 05m- mm; 3%. Eu.— N-m.v- E52- 5:55:00
55th 252 2555a 252 25:5; 252
:5;- 5525 23::

 

 

 

92.5280 2. 2...;

170

 

 

 

 

85 $5 25 $5 85 85 555 85 .3528: 5255.58
25 85 85 _55 55.5 85 85 85 8:83.282 2.5-"22>
85 85 85 S5 85 85 85 85 22> 82258.5 SE
85. 82- $5 25- 85 85 85 85 825.2228; 5282
humuﬂuv~ua3£b
5226‘ 3:283:
85 85 :2 85- 85 $5- 85 85 SEE 28. :22 95a
85 85 NE 85- 52 _ Z- 85 85 29:5 85 ___-.2 02S
85 85 82 £5 35 2 5 85 85 32:. 5655553525
85 85 52 3.5- 22 85- 85 85 SEE 22: :22 .55
85 85 $5 :5- 85 :5- 85 85 22.5 235 852
85 85 S2 85. :5 S5- 85 85 SEE 235 .989
85 85 $5 85 82 _55 85 85 222.5 5828 .8552
85 85 $5 85 85 85 85 85 332582 8258 E2
85 85 MS 85- 52 85 85 85 22> 3225 025
85 85 85 85- 52 .2 85- 85 85 ca: 2525 53:32
huﬁhmhuuuahaﬁnv gﬁuﬁﬂ 39¢:
85 85 85 85 85 E5 82 85 23.5 22: 82 new...
85 85 85 85 85. 85 m _ .N 85 23:5 222 82 005>
85 85 85 85 53- m:- 82 $5- 23.5 283553550
85 85 85 85 85 82 z: 85 23.5 255 =§< EC
85 85 85 85 25 85- 85 85 235 232 852
85 85 85 85 85 $2 a ._ 82 23.5 235 .925
85 85 85 85 :5 85 35 85 235 35528 28552
85 85 85 85 82 $5- 82 85 22> 25:5 025
85 85 85 85 85 85- 85 :5- cs: 23.: 3282
82.222886 838m 235
N @000 N .mooo N @000 N .mooo N .mooo N .mooo
m._> cﬂAonﬁm mu> cVoaﬁm mu> cﬂAogvm mu> cVosﬁm— mu> wﬂAosﬁm mu> ova—Um
:8: 225 23.5

 

 

 

.951— :32835 353:: .3 9.30 5.3! 955:5 EEO .8.— oﬂﬁ—U .3523 ac 3558.83:

Nah «Sun.

171

 

 

9».— wod Ewe mad- cm.~ aod cw; Cd mm; 3.0- wed mod- om-E moEEom 3.
end mod emd 86 3.0 cod mag 26. on; wed Nm.— 36. 8.3 832 32
34 Ed- am; cod omd 56- mmd mod- 2.6 86 3d mod- 3% 33:35 a
ma: cod- 3”.— ccd Ed 86- and 86- $6 36 36 mod :6 v.25: u
cmd mod vvd mod 3; cad- mod m _ .o 84 Ed mad cod _Eom h> e 325 .820 n
R; mod Rd 56- Q: cod- 36 cod Ed 36 cvd mod £52 .9 w $ch .520 u
we; 36- 26 $6 mod 26- an; mmd- and mad 2: and I: E .06—«Eon— 8w§o> *2
$4 3 .o- 2d mmd- 84m $6- and Cd- 26 56 mm; mvd- II E 332 how§o> n
hmd 36 3 ._ Ed- 36 Ed and 36. no; 3 .o end vmd- 8805 2050:8580
mad 86- 2 .c 8.0- wad «ﬁe mmd mad 2: m 2 .o- 86 cod 580? hemp—om ﬂ 3502
on. _ cod Q .N 86- mcd cod wad _od emﬁ 8.9- 8N mod- vuﬁoBHhoaaDoEaZOmm/x
vﬁo mod- 3. _ end mhé 36 um. m mud a _ .c 86- _ _4 omd vu_ﬁumoBOA-o=_m>aomm<
cad mad- cad c _ .o 9mm 3 .c and 26 av; _ _d- 36 8.9- 02852:
cod 86 EM— Nod- and cod :5 cad and 56 and 85 Amao>v om< Bang—A
RN- Nod- 2: 85- 34 mod cad mod 3." mod- cmd mod- 8% e HA 0:8 @533:
«Nd Ed mm; 36. Ed 56- m ﬁc 86- Cd 56- SN :6 ﬂuo> m-c 26m 9:385
mmd mod- mm.m SA. SA ovd no; 31.. 36 mod and mmd m~>m =< “coma/w @533:
36 cod cm._ 8.:- S.— Sd- cad mod and 86- 5.0 56- Eu on WA: 8532
de 86 _m._ 86 mad vod 36 36 emd wed cad _od :6 cm _ 8 a: .8502
m _ ._ _od- mud _od- wed 86 cm; mod- and 86- had cod ﬁEo-C om< .5502
cod cod wmd mod ch; 86 w; 0 HA 26m .5522
me; mod NM.“ mad :4 mod m30> m-o 0:8 3:8:
mad 56 ca; 86- Nad— eod- on.» 86- oN.o_ mod- v:- acd- vm-N mo: ow< EEO
22$ wmd cod 36 2mm :6 M34.- 36 a; 36 36 mod o; 8: ow< 220
34 2 .c E A :6 vﬁm 36 8N 36 $6 86 cm; mad 252 8 BED
humuﬁsuuuuuutb 383.33% 3.3 33%5

mu> equsum ﬁ> eVosvm m..> quosvm mu.» onsvm E> oquavm E> oVoacm

=5th 352.5 0:95

 

 

 

6822.85 3. 2.3.

172

 

 

 

 

 

8.888- 8. _ _ 8: 58232: 82
85 85 2 8:35
8588 32: 8.88 H83 23 8 .2352
_55 8.8 85 8. 2 N 85 8.2. :5 8.2 $5 8.2 85 85 :228
85 8.8 85 2.8 25 8.2 25 8.: 85 8.8 25 8.2 me25 88
85 8.8 85 8.2. 85 8.8 85 8.8 85 8.8 85 8.8 2532855
85 2:: 85 8. z: 85 8.82 85 8.5. 85 8.8 85 8.8 8.5.5.39
85 2.8 85 8.: 85 8.8 :5 8.2 85 8.8 85 8.2 828538 28:
m _ 5 8.8 85 2 5 85 2.8 85 8.2 85 8.8 85 88 238. 83>
85 28 85 8.8 85 .88 85 85 85 E8 85 8.8 225 2:38:
85 85 85 8.8 _55 :5 85 8.: 85 88 85 25 32...: 232
a: w: a: ma NC at a: m..— N: m: a: w: Unﬁm hug—.302
85 88: 85 8.8 85 8.22 85 8.8 85 852 85 2.8 a? 230
35823: 5. 3.8%.:
2 8 2. 2. 85 $5 85 8.8 85 n _ .8 85 8.2 82.3282 2::
85 8.8 85 8.2 85 :2 85 8.8 85 8.8 85 8.8 355225
85 8.: 85 $58 85 8.8 85 8.8 85 2.5 85 8.2 8:225 2; 2.5
85 8.8 85 a. _ m 85 8.2 85 85 85 8.2: 85 8. :. 8.588
87:20 @3525» .3 3.83—Ema 2:3. :8 358.85 «moh- ohaauw-EU
8._ 8.8- 82 8.2.- 8.8 8.8- 85 8. 2- 82 m _ .m- 85 8.8- 55 35250
85 85 85 85 85 85 85 S5- 85 85 25 85 8 .25 8358 8
8.8 85- 85 25 8.8 85- 85 25- 8.: 85- 85 85 8 55 832 2
85 85- 8: 25 8.: 85 85 85 85 85 85 85- 8-8 83:55 8
8.2 85- 85 85- 85 85 S5 25- 85 85- 85 25 8-8 832 2
m; 8qu85 5 826m m; 868-". m; 8228 m; 8qu58: m; 82.3
35 2.25 23.5

 

 

 

823885 3 03.8

173

 

 

 

 

2:- 85 8.8 $5 85 85 55.5 8:58: 88:58:
8.: _55 85 85 85 85 85 85.3882 3:8;
8.: 25 8.8 85 85 85 85 85 38:65 8:5
8.8 8. _- _ 8 8.8- 85 85 85 . :558285 62.37:
amuﬁhuuUQKQS-v swﬁm39~a~ Bﬂﬁmﬁgsh
85 85 2 .N 85- 85 85- 55.5 22:: x8: :§< :58
85 85 8.: :5- R5 85- 85 38E 8:: :§< 08>
85 55.5 $5 25 :8 85 55.5 3%.: 85893526
85 85 8.8 $5- 85 25. 55.5 22$ 85 :92 b5
85 85 85 m 2 5 8.: 8. _ - 85 033.: 23: 852
85 85 85 85- 8.: 8. _- 85 DEE 25. .28:
85 55.5 8.: 85 :5 55.5 85 22:5 883.: 88.2
55.5 85 85 85 85 85 55.5 323.8 828$ :5
85 85 2.: 85- 25 _55 85 :8» 38:: SE
85 85 8.8 85- 8.: 85- 85 as: 228 «9:82
unmuuwhuu9§h§~bv S535“ $3.:th
85 85 55.5 55.5 8.: $5 2.5 85- 238 85 :§< 85
85 85 85 85 8.8 85 8.8 $5 088 x8: :92 08>
85 55.5 85 85 8.8 8. _- 2.8 8. _- 0:95 8:333:80
85 85 85 85 8.:- 8: 8.: 85 238 85 :92 55
85 85 85 85 85 85 85 $5- 238 25: 8.32
85 85 85 85 8.8 88 85 85 0:8,. 5:2 .982
85 85 55.5 85 8.8 _55 8.8 _55 238 388.8 h8.82
55.5 85 85 85 85 85- 85 85 83 2:8 8:5
85 85 85 85 8.:- 85- 55 25- :5: 23.: 8302
8383239 883,: 9.85:
N @000 N @000 N @000 N .wooo N Moo-u N @000
3:28 an .28 0:28 1 8:28 an .28 0:28 .2 8:28 an .28 0:28 .2
=8: 82; 238

 

 

 

maﬁa—om «on: Sagasom no «53> 3 0.30 5135 955.50 330 .3.— oumoau 5:33: .3 3.3383qu

MS 03:8

174

 

 

 

 

dmd _dd nmd mod dfm mod mm; :d do; udd- wmd odd- dN-w_ 8.080”— a
dwd Nod hm; w_d wmd Nod- mm._ :d- and wdd wmd mod. 8.: 8:02 u
mNd _dd RA :d- hwd Ndd- 2d 5d- ddd odd mwd odd 2% 00—0805 n
ded mod and wdd- Ed 5d SN 3 d- N55 wdd dmd mod- 26 8.02 0%
SA 8d mm; add wmd Nod- mwd odd ww._ add dmd Ndd .805 ta. 0 .895 0020 u
cod Ndd we; odd wwd wdd- mad nod- mmd Nod wwd dd 032 0> e .8ch 0030 u
and wdd- mm; and- and Qd- om; mmd- 2.— 3d and :d- I: E 8.0805 h0w§o> w
an; :d- mm; de- Cd hwd- wmd :d- mmd mdd- owd Ed. 32 5 00—0—2 00w==o> u
mwd mdd no.5 de- whd add Sud cod _ _A m _ d _ _._ 3.? E0005 0050:6500
wed _dd w. .5 end. mwg :d mod n _ d- ddd ddd on; Nwd- :0an 8E0m m— 0050.2
mm. _ ddd mo. _ ddd ”Nd ddd m0_m000.~h0an-0=_0>u0mm<
de wdd dd. m 8d n md m _ d 0:83-00 _ -0=_m>u0mm<
m dd ddd wed odd- dad a 5 d be; dmd and 8d- dmd odd- 00.8.50:—
ﬁmd ddd cm; _dd- mod _dd wdd ddd 3d cod a; 5d Amum0>v 0w< 2:332
Nb.— Ndd- am; mdd- 5m; 5d 2d ddd 54 mod. mm; wdd- mCr e HA 26m @533:
IN udd- dwd Ndd ed; wdd M2.— hdd- mm.“ odd _wd mod m30> m-d 26m 283::
34 3.0. mm: mud- ww; mmd 2 ._ hmd- mwd 2d wmd h 5 d mu>~ =< E0mn< @5333
cod _dd bad 3.? de ddd mad Ndd- NNN mdd- ewd _dd- :8 dm _HAE 00502
mmd ddd cad 8d and mod dwN mod mod _dd wwd mdd 80 dm _ 8 a! 00502
dmd Ndd- 5N; _dd an; _dd- 5d 3d. 8N Ndd- an; Ndd- Am30>v 0w< 00502
and _dd- 34 mod wed mod 2d ddd cod _dd odd mdd 00> 0 HA 030m 00522
3d mod mm; mod RN d_d cmd mod wmd Nod 5d mod mB0> m-d 035 00502
and No. T and mm; mm; ood- nmg Nwd mw._ wad- and ww._ mh>m =< E0mn< 00502
cod ddd Nwd ddd wwd_ edd- Nod bdd- no.2 ddd- med wdd- wN-h 82 0w< EEO
3N5 Sud N—d wmd _w.m :d wad w~d Na; mod do; mod e; mow: 0w< EEO
Ed 2d add sod Ed 3d and dmd and mod wad 2d 222 mm EEO
numuumsuuuusuS-v Edﬁumaayt 33 3:33.55

3:20.5- Em .Bm 0:205. i 3:805- Em .EN 0:805. 1 3:205. Em .BN 0:20.5- i

:35. 0003.5 0:85

 

62.5253 2 2.3.

175

0.5055 bag“ 56 05550.5. :0 00502

 

 

 

 

 

5mddowm- wmdomd5- 56255500555 mod
odd 8d NM 05505
do.aww5m 05580.5. Emﬁﬁ ood5.wa 0:80F “05 Bo .50 50558552
Nod 05mm 50d 5%.: ddd nadm mod wodm mod dN.~N 5nd wwd5 550555535
odd 50.3 odd 5wd5 nod 555.2 add 3.2 cmd dmd wwd mad 00055 noon
dd d we. 5 m ddd waSw wod 55:55 m5d wc.w5 odd 55:.» wdd 3.5: 05303500555:
8 d adds ddd 3.85 ddd wm.md5 mod madw odd Edi 5od wwdm §§E=BD
mmd 3.2 05 d R a5 5od aﬁam 5 5d wndm wwd wm.m5 Ed 5w.w 58505855800 2055:
a: a: 0: a: a: a: a: 0: 0: a: a: a: 050mm< 0=5m>
odd 5:: and NON 05d 5.5km awd ww.5 w 5 d wwd omd 5w.N 0:55 0:80.35
mmd 35 Ed ow a N5 d 5m.w wod mmd mod md.m wdd wwd Ems: H0555052
5md cm 5 m5d 05w ddd No.05 Nwd dwd mod wwd mod mad 0&5 505582
odd 5.35 ddd 5a.wa ddd 3.2 5 odd modm ddd dmwm5 odd Sww 0w< E5550
§50<0§0- av 5:33.505
0: a: 0: end wad mod «NS ddd owda odd cmdm b555n0550>< wED
ddd aw.mm ddd m5..5m wwd wwd mmd 0N.w odd 502 3d om.~5 5055:8005
odd no.2 odd a5 dm wed 5d.m cmd mod ddd mm. 5 N .5-5d wwd 0053050 5:50 00-.5
Odd cm. 5 m odd waw 5Nd 5 5.m5 Ed 348 8d made dod awdm bEuuk
25.8.59 055505.55» .3 00.305.55.55 5:505. .85 0.053550% 500.5. Egg-5.50
mm 5 mmd- wn.5 wn w- 595 com. ~53” Qua- mm.5 cod- 3.5 wwd- E505. 5:80:00
wwd mod mod :3 wwd mod mod 5d d 5 d Nod- wad wwd do h0>O 0050805 55
Nw.5 Ed- wwd add- 5: .m wwd- Nwd 5 5d mwd odd- dmd m5d- dd 50>O 005052 55
5-wd mod 35 Ed. 35 8d Nod odd mod cod Sd add 8-5m 85050-5 55
odd add- wad add add ddd am.5 m5d Rd 5.0. wdd odd do; N 005052 55
00:80.5. 2m .2m 0:80.5- :5 00550509 Em mam 0:80.5- :5 00:050h em .EN 0:209 :5
5500:. 05035 0535

 

 

 

55.2.5an 2 2.3

176

 

 

 

 

 

 

 

 

00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 6.000002 0008:0000
00.0 00.0 :2 00.0- 0.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 0030350002 8000;
00.0 00.0 2 .0 00.0- 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00> 30000020 8:0
00.0 00.0- 00.0 m 0 .m- 0.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 5800200005 0282
humuuwsuuuuhaﬂb LQVESLN BEﬁBL-N
00.0 00.0 00.0 00.0 2 00.0- Ea 2.0- 00.0 00.0 00.0 00.0 2200 00.0 0.3. 0500
00.0 00.0 00.0 00.0 :2 3.0- a: 00.0- 00.0 00.0 00.0 00.0 2200 x05 09;. 80>
00.0 00.0 00.0 00.0 0.0.0 00.0 00.0 00.0- 00.0 00.0 00.0 00.0 00000 00502935000
00.0 00.0 00.0 00.0 S 00.0. 00.0 00.0- 00.0 00.0 00.0 00.0 3200 x05 00% .55
00.0 00.0 00.0 00.0 2.0 R0- :0 00.0 00.0 00.0 00.0 00.0 3200 005. 02:2
00.0 00.0 00.0 00.0 00.0 00.0- 00.0 3.0. 00.0 00.0 00.0 00.0 22.0 0000 .280
00.0 00.0 00.0 00.0 00.0 00.0 02 00.0 00.0 00.0 00.0 00.0 32.0 200.000 30.002
00.0 00.0 00.0 00.0 3.0 3.0 00.0 00.0 00.0 00.0 00.0 00.0 507080 $008.0 E0
00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0- 00.0 00.0 00.0 00.0 00> 2200 8000
00.0 00.0 00.0 00.0 00.5 00.0- 00.0 2 .0- 00.0 00.0 00.0 00.0 550 300.0 08.82
homuhmkm~9§s§£0 émﬁuuk 33‘:
00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 2.0 00.0 00.0 00.0 0300 x05 00% 0:30
00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0. 00.0 00.0 R0 .0300 00.0 002 80>
00.0 00.0 00.0 00.0 0.0 00.0 00.0 00.0 03 03- 00.0 _0._- 0300 00500535000
00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 0:. 00.5 00.0 00.0 00000 00.0 002 000
00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 3.0 3 .0 00.0 00.0 0300 0090 852
00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 02 _0._ 0300 .05. .983
00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 03 00.0 02 _00 0300 3:08.00 .3802
00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00> 00000 800
00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0- 00.0 00.0- 5000 0300 0232
8.050003305
0.58-.0 0.00:0
N @000 N .-«000 N .mooo N H000 N @000 N .moOo
.82 0.0.0 .82 0-5 .82 0.0.0 .82 0; .82 00-0 .82 0-0
_05 39000 0300
2.0.02 .0 03 00.6 E 25 008: 20.25 0000 .80 800.6 00080 .0 3:200:35 0.0 0.000

177

 

 

 

 

we.5 wed edd eed e5.5 med- Ed 8d. wwd Ned w5d 5ed em-w5 005035 w
de Ned- e55 wed- 5nd med- awd Ned wwd Ned 55d 3.? M56 00508005 50
Ed 3d. 3.5 aed Ed Ned- 5ed eed NW5 aed 5nd Ned- mTe 005052 w
wwd wed ”5d 5d 8d eed end wed- :5 wed and med 580-5 .5 e 0.05ch 005050 w
ew.5 med a 5 .5 wed- wwd med- 9.5 aed.- ac.5 wed wa.5 aed- 05052 0> c 0050503 005050 w
35 55d- aed mad- wN.5 3.? and wwd- wed e5d mNd 05d 5.555 :5 0050:0005 00wcao> w
5m.m a5d- mad wmd 3.5” and- mad ”5d mmd med- mmd de 5.5: :5 005052 00w==o> 00
2 d 3.? end m 5 d- end aed w5d Ned eNd med amd med 05.000005 00058850500
wmd med- N95 ewd- owd med 3.5 end wwd hed- w5.5 NNd- 50:83 005:0m 05 00550035
wa. 5 eed ew. 5 5 ed- 5m .5 eed me. 5 eed and 5d. 0 5 .5 eed cEHOBH00aeD-os5w>«omm<
”Nd med cod 8.? www Gd 5.56 and med 8.? mad m5d 50:55.5.500304-o=50>0000<
med eed wad aed- 59m mad e5.5 aed mad aed- e95 w 5d- 00:00:05
85 3d. ”5d eed mad 5ed a5d eed wwd 8d and 5d 0080C 0w< 50:30:55
end med- 5.5 wed- ”Nd eed and wed mm.5 med- wed med- 00> e HA 26m 5053035
3.5 wed- emd 5ed mad 5.? N05 aed emd med mod m 5d 000»; We 26m 5053035
med med.- Ned emd- mad end mm.5 emd mmd N5 d wad emd m0>~ 55< 80055450 50:30.35
mwd eed had med. 5 5d eed 5m.5 mod. and med- mad 3.? Eu em WA“: 005035
mwd ded aed 5ed 5w.N med 35.5 med emd med 5nd cod :8 em5 8 055 0005835
aad 5ed- ww.5 Ned- w 5 .5 8d. 3 .5 8d. med Ned- wed Nod- Guava 0w< 005035
5~d eed 3.5 wed- cod med dad 5ed Nw.5 Ned med 5ed 00> 0 MA 0:35 0255052
aN.5 med 3: Ed w5.N wed «W5 wed wwd Ned med aed 0000> We 26m 005052
5 m.5 and- 3.5 mm. 5 - de m 5 d- 00>N 55¢. 0:00“?- 005035
mad 5d. and med- wwd5 wed- wm-a 002 0w< 5055550
mad mmd mmd aed 3.5 med v-5 mow/5 ow< 5055.50

3.5 5 5d 3.5 eNd MNN 05d mm.m de mmd med wad and 05052 05 5055550
uumuﬂsuuuauutb 323.33% 35» 3:33.35

N «000 N “5000 N “5000 N H5000 N “5000 N “5000
.0035 wN-e .0035 0-5 .0035 wN-a .0052 9.5 .0052 wmé- .0035 95
55000-5. 05055005 0:05:05

 

 

 

500000080 3 0.00.5.

178

.0025 been 56 030-5. :0 0082

 

 

 

 

 

 

5madmaam- wand”:- 5000555585515 mod
memed amaed NM 0550005
emwwN HwNA mmwe “02 e5 0050 do 5.58:2
eed N5 .ww eed 5a.aN eed mwdm eed madn 05d Ne.w5 eed eNdm 550%535
Ned e5..a5 Rd end eed am.5-N aed mwd mwd med NNd ea.5 5 0005.5 woo-.5
eed wade 5e.e m5.wN eed medw m5d me.w5 eed 5mdw eed wadw 0030:5000ec5
eed 5w.wa5 eed 5w.w5- eed 55.65 5 eed we.ae5 eed eN.Nw5 eed ma.ma 355555—80
:d wa.w5 mod aw.5 5 med mm.mm mmd we.: 5nd 3.: ead 50.5. 5005:0er00 50505555
NNd we.m amd med eed Gem eed am.5-N med Nah we as. 5 m~0mm< 0:50>
eed e5.m5 Cd emm wad N5d ee e 3.3 ewd e555 eed 35 5 ozum 50:80.3:
wwd ewd 50d 85 med wwd 05 d aem med mms wed ead 2&6: 00555052
amd aa.5 aed end eed mm.e5 N5 .e 5m.w mwd we.m Ned wwd 0:50m5 0055635
0: a: 0: a: a: m: 0: 0: a: a: a: a: 0w< 5055550
Eeﬁmaei av 53.32%:
a: m: a: a: med 5 e5 eed e5 a eed 5...: eed mawe b5550550>< wEQ
eed w5.mw 5d 0 5 .e5 wad am.5 wmd ewm eed ewmw eed w5.m5 5050:8095
eed ewdm 5e.e ema wm e wwd 5e.e mna eed 5 5.m5 eed wmm5 0050350 50:0 00-5
eed aedm eed wmd5 5e.e mwdm 5e.e aw.5m eed am. 55 eed wwdn bmmuuk
0.5.8.50 05050505» .3 oouuou5=u5m 5:53. .55 353-535 «00.5- 005.5%-5050
wmd awd eed aw.N- eed Sum- wa.5 em.m- mwd mad- wwd am. 5- E005. 5:80:00
end see 85 mwd- NNd Ned wed aed mwd eed amd aed- ee 00>O m0505005 05
50.5 w5d- w5.e med- aa 5 mmd- wm e aed- mwd aed- awd 55d- ee 00>O 005035 n
mwd Ned am.5 m5.e- wa 5 aed we.5 eed awd med m5.5 med- ee-5N 85080» w
wm.5 med- eed 5e.e mm e Ned awd med N5d 5ed- aed 5e.e- ee-5N 0.05035 u
we med Ed 3d. 3: Ned Lamd e5 e aa.5 aed- wad med- eN-w5 0050:0005 05
N wooo N ..50oo N @000 N .5000 N “5000 N “5000
852 wmé mow/5 0-5 .0052 me .0052 v-5 .0055 wwd mow/5 e5
5500; 0002005 0:05:05

 

 

 

58200.80 3 2.3

179

APPENDIX B

Derivation of the Estimated Probabilities and
the Unconditional Marginal Effects
for Nested Multinomial Logit Model

180

APPENDIX B

Derivation of Nested Multinomial Logit Probabilities and
Unconditional Marginal Effects of Regressors

W

To derive the choice probabilities for the nested multinomial logit model, ﬁrst write the
single-level multinomial logit model in nested form. The single level model in its simple
form is written as

PJ- = exp(Vj)/): [exp(Vk)] for k = 1 to J

If we now assume that the tree has the following two-level structure,

Visit No Visit (Limb= L)
| | |
Public Private Traditional (Branch = C)

 

we can rewrite the probability as a function of these two levels of the decision, L and C
Pc/ : exp(vc|l + V ! yz/ 2c 6XP(VCI- + V!)

= (Pct ) * (Pe)

Pa
Pa =Pc|l *1): = [CXPWcu iii/2c e)'1P(ch)]"‘ [eXPWc V24 exp(V- )]*_[Zg expW-u )JDZ- CXPCV: )]
[2! 2c €XP(Vcn + V l )1

Deﬁne the inclusive value parameter as the log of the denominator of the conditional
choice

I ( = 1n 2c exp(vcll )

SO

Pcl! : [exp(vclf :I/Xc exp(Vc|I )]

and

181

 

P, = [exp(V, + ﬂy}; exp(V, + tI)]

and

PC, = [exp(Vc" II/Xc exp(Vc,,)] * [exp(V, + tI)/):, exp(V, + tI)]

If, however, the parameter of the inclusive value, 1, equals one, then
Pd = exp(Vcl, + V, )/Z, Z, exp(Vcl, + V ,)

which simpliﬁes to our original multinomial model with k choices

Pj = exp(V,-V2 [exp(V-J] for k = 1 to J

M E S

In a two-level model, starting with the facility choice limb, we deﬁned the probability of
each choice as

Pcl! = [exp(Vcll /Zc exp(Vcll )]

There are three alternatives at this level, public facility, private facility, and traditional
facility. If we refer to each indirect utility function using the notation B“) for the set of
estimated coefﬁcients and X as the set of regressors, we can write the probabilities as

P0) = exp(B‘” X) / exp(B‘” X) + exp(B‘Z’X) + exp(B‘3’X) = CXP(B“’ X)/ZeXP(B‘°’X)
P(2) =CXP(B(2) X) / eMidi") X) + CXP(B‘2’X) + CXP(B‘3’X) = ewi‘z’ X)/XeXp(B‘°’X)
P(3) =eXP(l3‘3’ X) / “1903‘” X) + exp(B‘z’X) + 6Xp(l3‘3’X) = 6Xp(l3‘3’ X)/XCXP(B‘°’X)

Let alternative three be the alternative we normalize to zero.

To derive the marginal effects for this conditional choice, we differentiate each equation
using the quotient rule

d g = [v"'u'-u"‘v']/v2
dx v

182

 

Therefore, for the conditional choices, the own-marginal effect of a change in one of the
facility characteristic, say of alternative (1), equals

M1) = ZQXD(§(°)X)*h1-§2£D(Dm X) _ QZSIKQ") 29:12“ QKDCQU) X)
dxn [ZCXP(B(‘”X)]2
= h“ (I) (C) _ (l)
[ZCXP(B(°)X)]2
= inPw (I‘Pnz)

For the marginal choice between the limbs of whether to have a health care visit or not,
the probability is given as

P, = [exp(V, + rI)/£, exp(V, + tI)]

Using a“ ) to refer to the set of estimated coefficients and X as the set of regressors, we
can write the probability of having a health care visit as

P, = [exp(a“ )X + 1:1)/£- exp(oc‘l )X + 171)] where I = log [Zcexp(B(°’X)]
Because the facility characteristics do not enter into this level of the tree, they have no

estimated parameters in this level. However, each own facility characteristic enters P, via
1, so we must still differentiate this equation.

£1.11: TPI(1'P1)

CH

and

ﬂ = hum (I) = bliPclt
an Zcexp(l3‘°’x)

SOPl'zbliPller!(1'Pl)

With this and with the marginal effects on the conditional probabilities we can obtain the
total own-unconditional marginal effect for a change in a facility characteristic

183

dﬁcyp = Pc|l ' P! + Pelt Pl '
dxc,
= bci Pen (I’Pcy )1)! + Pcv bli Pep 1: Pl (1 ' Pl )

= bci Pcl! Pl [(I'Pcll ) + Pcll 1"(1 ' Pl )]
Tracing through a similar process as outlined above, one can obtain the unconditional

cross-marginal effects for changes in the characteristics of other health care alternatives

ﬁling/‘5 = bci I)c|t Pl Pnl! [AL-(1 ' Pl ) " 1]
dxci

and the unconditional marginal effects for changes in non-health facility characteristics,
such as individual, household, or community attributes

ﬂe- : Pelt Pl [bci(1'Pcll) + (“I +1 Pcl! bci (1 ' Pt )]
dx,

184

 

APPENDIX C

Supplementary Tables

185

Table C1. Community Inﬂuences for Any Child Curative Care Visit

 

 

coef. z-stat
High Elevation (0-1) -O.22 -2.16
Island (0-1) 0.06 0.92
CommAngiped/Pumthr -O.62 -0.76
CommAnganitGarstposl -O.54 -1 .91
CommFrquaterShortages -0. 12 -1 .65
CommHasImprovedRoads -0. l 5 -2.39
CommunityHasBank 0.01 0.15
CommAngridgeOwners 3 .09 1 .90
ComdenTimeToWaterSource -0.05 -3 .26
CommAngodemToilet -0.58 -2.53
ComPrice-Current 0.00 -0.5 l
ComPrice-Lag 2 Mos. 0.00 -1.16
ComPrice-Lag 4 Mos. 0.00 1.30
VegOilPrice-Current -0.07 - l .22
VegOilPrice-Lag 2 Mos. 0.11 2.01
VegOilPrice-Lag 4 Mos. 0.14 2.30
IanormulaPrice-Current -0.04 -1 .27
IanormulaPrice-Lag 2 Mos. -0.11 -3.83
IanormulaPrice-Lag 4 Mos. -0.09 -3.31
Rainfall-Current 0.00 -O.98
Rainfall-Lag 2 Mos. 0.00 0.35
Rainfall-Lag 4 Mos. 0.00 1.08
Rainfall-Lag 6 Mos. 0.00 -l .45
Rainfall-Lag 8 Mos. 0.00 -3.12
Rainfall-Lag 10 Mos. 0.00 -l .55
Rainfall-Lag 12 Mos. 0.00 0.44
Rainfall-Lag l4 Mos. 0.00 -O.17
Rainfall-Lag l6 Mos. 0.00 -2.79
Rainfall-Lag 18 Mos. 0.00 1.58
Rainfall-Lag 20 Mos. 0.00 -1.19
Rainfall-Lag 22 Mos. 0.00 -2.37
Rainfall-Lag 24 Mos. 0.00 -O.65
Constant -5.41 3.09

 

 

Notes: 1. Main results from this regression, the facility and indvidual- and

household-level effects, are presented in Table 6.1. 2. Outcome of no visit is the

comparison group. 3. Regression standard errors adjusted for clustering on

individual id.

186

Table C2.

Pooled Sample — Complete Results for Baseline Flexible Speciﬁcation of

Determinants of Facility Choice for Child Curative Care Visit

 

 

 

Public vs None Private vs None Tradl vs None

coef. z-stat coef. z-stat coef. z-stat
Dist Nearest Public(km) -0.360 -4.46
Price Public Visit 0.092 0.31
Number Personnel Public 0.006 4.55
Doctor Ratio Public 1.544 3.41
Nurse Ratio Public 0.1 16 0.38
ORT Available Public 1.042 4.72
OthrDiarDrug Avl Public -1.221 -4.16
Vacc Avail Indx Public 0.615 4.37
Famp Avail Indx Public 0.335 1.13 . .
Dist Nearst Private(km) -0.069 -2.42
Price Private Visit -0.01 7 -0.81
Prvt Provides PostNatal 0.439 1.11
Number Personnl Private 0.002 1.25
Doctor Ratio Private -0.400 -1.05
Nurse Ratio Private -0.230 -0.52
ORT Available Private -0.403 -2.00
OthrDiarDrug Avl Privat 0.305 2.12
Vacc Avail Indx Private -0.757 -1.92
Famp Avail Indx Private -0.601 -2.09 . .
Dist Nearst Traditl(km) -1.990 -5.41
Price Traditional Visit 0.241 2.10
YrsEduc MedianTrlervdr 0.010 1.31
F ormlTraing MedianTrdl . . . . 0.466 6.42
Child is Male 0.098 1.34 0.200 3.08 0.122 2.20
Child Age Mos 1-6 0.042 2.15 0.121 6.72 0.372 15.15
Child Age Mos 7-24 -0.082 -12.66 -0.063 -12.52 -0.002 -0.33
Mother Absent All 2Yrs -0.227 -0.43 -0.206 -0.36 -0.924 -1 .57
Mother Educ 0-5 Years 0.039 1.17 0.083 2.41 0.039 1.55
Mother Educ >= 6 Yrs 0.020 1.48 0.028 2.70 -0.002 -0.20
Mother Age (Years) 0021 -2.51 -0.009 -1.28 -0.009 -1.42
Mother Ht to 150 cm 0.023 1.82 0.030 2.71 0.004 0.46
Mother Ht>=150 cm -0.021 -2.04 -0.004 -0.46 -0.004 -0.62
Husband AbsentAllZYrs 0.194 0.68 0.251 0.99 -0.575 -2.55
Husband Educ 0-5 Years 0.054 1.56 0.006 0.17 -0.032 -1.35
Husband Educ >= 6 Yrs -0.029 -2.22 0.012 1.23 -0.025 -2.74
Husband Age (Years) 0.008 1.19 0.005 0.69 -0.005 -1.04
Insurance -0.087 -1.27 0.192 3 .42 -0.016 -0.30
AssetValue-LowestThird 0.018 0.14 0.667 5.89 0.034 0.36
AssetValue-UpperTonhrd -0.005 -2.38 0.001 1 .04 -0.002 -1 .99
Mother Is Senior Woman -0. 121 -0.96 0.119 1.15 -0.060 -0.65
Grandmother Present 0.045 0.37 0.067 0.69 -0.019 -0.20
# Younger Males -0.063 -0.47 -0.365 -3.04 -0.195 -2.32
# Younger Females 0.092 0.62 -0.l84 -1.44 -0.120 -1.29

 

 

187

 

Table C2. (continued)

 

 

 

 

 

Public vs None Private vs None Tradl vs None

coef. z-stat coef. z-stat coef. z-stat
# Older Under 6 Yr Male 0.026 0.66 -0.045 -1.28 0.029 0.95
# Older Under 6 Yr Feml 0.066 1.66 -0.012 -0.34 0.067 2.16
# Males 6-13 0.043 1.13 -0.019 -0.57 0.006 0.20
# Females 6-13 0.011 0.29 -0.014 -0.44 -0.020 -0.70
# Males 14-20 0.022 0.50 -0.034 -0.94 0.036 1.03
# Females 14-20 -0.078 -1.74 0.091 2.76 0.019 0.59
# Males 21-60 -0.012 -0.24 0.019 0.47 -0.046 -1.29
# Females 21-60 0.012 0.26 0.071 2.02 0.000 -0.01
# Males Over 60 -0.077 -0.61 -0.l89 -l .82 -0.168 -1.66
# Females Over 60 0.018 0.14 0.036 0.38 0.044 0.50
High Elevation -0.946 -4.83 -0.241 -1.31 0.012 0.08
Island 0.673 3.88 0.284 0.75 0.236 2.15
ComAvPiped/PumthrHous -2.1 17 -1.30 1.321 1.02 -0.849 -0.75
CommAvSantGarbDisposal -1.089 -1.96 0.250 0.61 -1.900 -5 .32
CommFrquaterShortages 0.395 2.53 0.022 0. 15 -0.428 -4.52
CommHasImprovedRoads -0.77 7 -4.94 -0.048 -0.41 -0.023 -0.3 1
CommunityHasBank -0.285 -2.08 0.081 0.52 0.277 2.70
CommAngridgeOwners 6.714 1.92 -1.016 -0.3 8 6.186 2.65
ComdenTimeTthrSorce -0.124 -3.49 -0.085 -3.41 -0.057 -3. 12
CommAngodemToilet -0.165 -0.33 -0.567 -1.79 -1.704 -4.59
ComPrice-Present 0.001 1.27 0.000 -0.21 0.000 0.25
ComPrice-Lag 2 Mos. -0.001 -0.58 -0.001 -1.63 0.001 1.51
ComPrice-Lag 4 Mos. 0.001 0.92 -0.001 -1 .30 0.003 3.76
VegOilPrc-Present -0.075 -0.79 -0.1 19 -1.45 -0.143 -1.91
VegOilPrc-Lag 2 Mos. -0.026 -0.29 0.160 2.07 0.094 1.32
VegOilPrc-Lag 4 Mos. -0.031 -0.32 0.206 2.45 0.131 1.72
IanormulaPr-Present -0.020 -0.36 0.020 0.43 0.003 0.08
IanormulaPr-Lag 2 Mos. -0.104 -2.04 -0.063 -1.42 -0.057 -1.40
IanormulaPr-Lag 4 Mos. -0.080 -1.70 -0.050 -1.23 -0.061 -1.65
Rainfall-Present 0.000 0.68 0.000 -0.25 0.000 -1 .12
Rainfall-Lag 2 Mos. 0.000 1.35 0.000 0.69 0.000 -0.44
Rainfall-Lag 4 Mos. 0.000 1.64 0.000 1.11 0.000 1.60
Rainfall-Lag 6 Mos. 0.000 0.50 0.000 -1.09 0.000 0.72
Rainfall-Lag 8 Mos. 0.000 -0.27 -0.001 -3.78 0.000 0.02
Rainfall-Lag 10 Mos. 0.000 -0.36 -0.001 -2.26 0.000 1.12
Rainfall-Lag 12 Mos. 0.000 1.04 0.000 -0.99 0.000 2.03
Rainfall-Lag 14 Mos. 0.000 0.66 0.000 -1.26 0.000 1.20
Rainfall-Lag 16 Mos. 0.000 -0.02 0.000 -1.70 0.000 -1.83
Rainfall-Lag l8 Mos. 0.001 2.14 0.000 -0.28 0.000 2.17
Rainfall-Lag 20 Mos. 0.000 0.64 -0.001 -2.55 0.000 0.89
Rainfall-Lag 22 Mos. 0.000 -0.94 -0.001 -2.25 0.000 -0.33
Rainfall-Lag 24 Mos. 0.000 0.50 -0.001 -2.77 0.001 2.81
Constant -4.514 -2.15 -5.482 -2.93 -2.836 -1.82

 

 

Number of Obs: 30919

188

Log Likelihood: -35349.89

Pseudo R2: 0.0636

 

 

Table C3. Nested Multinomial Logit — Facility Choice for Child Curative Care

 

 

Variable Public v Iliad! Pn’vate v Iragll Visit v None
Coef. z Coef. z Coef. z

InclusiveValue 0.359 3.46
DistNearestPublic(km) -0.252 2.79

PricePublicVisit 0.248 0.77

NumberPersonnelPublic 0.008 5.08

DoctorRatioPublic 1 . 144 2.19

NurseRatioPublic -0.461 1 .30

ORTAvailablePublic 1.019 4.18

OthrDiarDrugAleublic -1 .780 5.37

VaccAvailIndeublic 0.463 2.86

F ampAvailIndeublic 0.407 1.22

DistNearstPrivate(km) -0.01 1 0.31

PricePrivateVisit 0.008 0.24

PrvtProvidesPostNatal 0.849 2.00
NumberPersonanrivate 0.001 0.63

DoctorRatioPrivate -0.413 0.85

NurseRatioPrivate -0. 142 0.24

ORTAvailablePrivate -0.548 2.30
OthrDiarDrugAlerivat 0.474 2.59

VaccAvailInderivate -0.824 1 .68

FampAvailInderivate -0.255 0.75

DistNearstTraditl(km) 1 .895 3 .72 1 .895 3.72

PriceTraditionalVisit -0.065 0.41 -0.065 0.41
YrsEducMedianTrlervdr 0.008 0.69 0.008 0.69
FormlTraingMedianTrdl -0.742 5.09 -0.742 5.09

ChildisMale -0.077 0.93 0.048 0.63 0.145 3.30
ChildAgeMosl-6 -0.317 10.94 -0.242 8.53 0.217 6.78
ChildAgeMos7-24 -0.084 11.46 -0.065 10.66 -0.024 4.00
MotherAbsentAllZYrs 0.739 1.11 0.737 1.80 -0.598 0.30
MotherEducO-SYears -0.003 0.08 0.032 0.85 0.046 1.92
MotherEduc>=6Yrs 0.024 1.67 0.037 3.14 0.007 0.88
MotherAge(Years) -0.012 1.34 -0.002 0.24 -0.010 2.00
MotherHttolSOem 0.022 1.63 0.027 2.14 0.010 1.11
MotherHt>=150cm -0.012 1.02 0.004 0.46 -0.007 1.17

Table C3. (continued)

 

 

Variable Public x I rad] Erin/alum Visit v None
Coef. z Coef. z Coef. z

HusbandAbsentAllZYrs 0.818 2.50 0.897 3.12 -0.316 1.75
HusbandEducO-S Years 0.096 2.52 0.051 1 .35 -0.020 0.87
HusbandEduc>=6Yrs -0.001 0.10 0.038 3.29 -0.016 2.00
HusbandAge(Years) 0.015 2.21 0.013 1.83 -0.002 0.50
Insurance -0.026 0.34 0.234 3.64 0.015 0.37
AssetValue-LowestThird ~0.03 l 0.23 0.632 5.03 0.153 1 .82
AssetValue-UpperTonhrd—0.002 0.96 0.003 1 .97 -0.001 0.00
MotherIsSeniorWoman -0.106 0.75 0.162 1.41 -0.025 0.33
GrandmotherPresent 0.065 0.45 0.089 0.81 0.005 0.06
#YoungerMalesInHH 0.096 0.68 -0.231 1 .83 -O.212 3.12
#YoungerFemalesinHH 0.182 1.18 -0.105 0.78 -0.1 18 1.57
#OlderUnder6YrMale 0.021 0.45 -0.064 1.51 0.008 0.33
#OlderUnder6YrFeml 0.009 0.20 -0.072 1 .71 0.046 1 .77
#Males6- 1 3 0.042 0.97 -0.024 0.66 0.004 0.16
#Females6-1 3 0.029 0.69 0.002 0.07 -0.013 0.62
#Malesl4-20 -0.026 0.53 -0.081 1.88 0.018 0.64
#Females 1 4-20 -0. 100 1.90 0.070 1.79 0.021 0.88
#Males2 1-60 0.033 0.60 0.060 1.27 -0.025 0.83
#Female521-60 0.013 0.25 0.088 1.93 0.021 0.72
#MalesOver60 0.068 0.47 -0.016 0.14 -0.147 2.07
#FemalesOver60 0.000 0.00 0.021 0.19 0.021 0.30
ConstantTerm -0.644 0.26 -3 .384 1 .52 -2.207 1 .62
Number of obs = 15207 15207 30919
Pseudo R2 = 0.1177 0.1177 0.0209
Log Likelihood = -14402.588 -14402.588 -20980.033

 

Notes: 1. Standard errors in marginal visit choice column are cluster bootstrapped on 500
replications. 2. Standard errors in conditional facility choice columns are adjusted for clustering
on id. Note 3 on Table 6.1 also applies.

 

190

Table C4.

Full Cross-Effects Speciﬁcation

Pooled Sample — Facility Price and Quality Results for

 

 

 

 

 

 

Public vs None Private vs None Tradl vs None
coef z-stat coef z-stat coef z-stat

Public Characteristics

Dist Nearest Public(km) -0.291 -3.39 -0.188 -2.081 0.051 0.83
Price Public Visit 1.083 2.08 -1.269 -2.89 1.542 4.38
Number Personnel Public 0.000 0.1 1 -0.005 -2.21 —0.004 -2.59
Doctor Ratio Public 2.320 2.74 0.213 0.28 0.813 1.34
Nurse Ratio Public -0.198 -0.45 -0.093 -0.21 0.595 1.79
ORT Available Public 1.137 3.76 0.228 0.85 0.253 1.17
OthrDiarDrug Avl Public -0.923 -1.70 0.188 0.36 0.787 2.09
Vacc Avail Indx Public 0.464 1.90 0.462 2.16 0.444 2.74
Famp Avail Indx Public 0.940 2.16 -0.503 -1.37 -0.610 -2.01
Private Characteristics

Dist Nearst Private(km) -0.170 -3.1 1 -0.089 -2.24 -0.091 -2.94
Price Private Visit 0.010 0.10 0.053 0.66 0.009 0.15
Prvt Provides PostNatal -2.774 -4.07 -0.676 -1.17 1.043 2.51
Number Personnl Private -0.003 -0.90 0.006 2.13 0.008 3.56
Doctor Ratio Private 1.228 1.31 1.778 2.25 0.897 1.46
Nurse Ratio Private 0.449 0.44 1.329 1.57 1.825 2.77
ORT Available Private 0.596 1.46 -0.393 -1.27 -0.232 -0.93
OthrDiarDrug Avl Privat 0.232 0.80 0.742 3.00 -0.140 -0.61
Vacc Avail Indx Private 1.693 1.74 -0.924 -1 .24 -1.004 -1.65
Famp Avail Indx Private 1.213 2.00 0.027 0.06 -0.378 -1.05
Traditional Characteristics

Dist Nearst Traditl(km) 0.558 0.45 -1.943 -1 .97 -0.403 -0.51
Price Traditional Visit -0.390 -1.54 0.177 0.82 0.166 0.99
YrsEduc MedianTrlervdr 0.057 2.94 -0.055 -3.41 0.029 2.22
FormlTraing MedianTrdl -0.742 -2.49 —0.125 -0.45 0.175 0.69
Constant -4.136 -1.45 -5 .851 -2.18 -9.923 -4.68

 

Notes on Table 6.1 apply here.

191

 

“1111111111111i: