Demand for Child Health in Brazil: A Household Decision Model

fipe
Fundação Instituto de
Pesquisas Econômicas
Seminar on Child Health, Poverty and the Role of Public Policies
Interamerican Development Bank, Washington DC.
Paper number 2
Demand for Child Health in Brazil:
A Household Decision Model
Denisard Alves
dcoalves@usp.br
Walter Belluzzo
belluzzo@usp.br
February 2004
Demand for Child Health in Brazil:
A Household Decision Model
Denisard Alves∗
dcoalves@usp.br
Walter Belluzzo
belluzzo@usp.br
1
Introduction
This paper examines child health, poverty and the role of social policies in Brazil.
Demand for child health is employed as the base for analyzing the impact of
poverty on child health. The role of public policies will be discussed within the
framework of child health demand.
The data used in the estimations come from the Standard of Living Survey of
1996/97 carried out by IBGE, the Brazilian census bureau. This survey, which
includes individual anthropometric measurements, covered the Northeast and the
Southeast regions of Brazil. The former is known for its low living standards,
whereas the latter is much more prosperous, having been the birthplace of late
19th and early 20th century Brazilian industrial growth.
The Northeast and the Southeast regions each concentrate around 30% of
Brazilian population, but the Northeast accounts for only 22% of Brazilian GDP
while the Southeast contributes with 55% of GDP. The survey therefore covers
two strikingly different regions in terms of economic performance; this allows one
- after controlling for socioeconomic factors- to understand the role played by
∗
The author acknowledges the help of Fabiana Tito, Márcio Diniz, and Daniel Monte as
research assistants. Doracy Brisola supplied the secretarial assistance. Nemesis, a Pronex
Project of CNPq and Fundação Instituto de Estudos Econômicos (FIPE) contributed with
financial support.
1
history and geography in explaining the differences in demand for child health
between these two regions.
The paper is organized as follows. In the second section presents a model
of household decision process to derive the reduced form equation for the child
health demand. The third section presents the description of the data used in the
estimation of the child health demand. The fourth section presents the methodology to estimate the opportunity cost of time that will be used in the estimation of
the demand for child health. The fifth section present and discuss the estimation
results. The last section, conclusion, presents the policy actions to improve child
health in Brazil.
2
A Model of Demand for Child Health
Becker (1981) suggested modeling the household allocation of the full household
income – which consists of the available time and money income – as the solution
of a joint utility maximization problem. This framework can be used to define
the demand for health by defining a production function that transforms leisure
and other goods in health, which is included in the utility function.
Assume that there are m members in the household and J goods and let Cji
represent the total quantity of good j consumed by the members of the household
i. Then the household utility function can be written as
U = U (H c , H p , C, Lp )
(1)
Pm
where C = (C1 , . . . , CJ ) with Cj = i=1 Cji , j = 1 . . . J, and i = 1 . . . m. Positive
utility is derived from C, from the health status of the household children H c ,
from the health status of the parents H p , and from the leisure time spent by the
parents Lp .
Let the health status of the household, H, be a composite index of H p and
H c – the health of the household parents and children, respectively – and assume
it is given by production function like
H = H(C, K, Lp , X, u),
(2)
where K represents health inputs which affect utility only through their impact
on H, X is a vector of exogenous variables, such as household and family characteristics, and u are unobservable random factors.
2
The household faces a full income restriction. Full income consists of the
available time and money income given by:
J
X
pcj
Cj +
j=1
S
X
pks Ks = w(T p − Lp ) + Y
(3)
pks Ks + wLp = wT p + Y = Y ∗ ,
(4)
s=J+1
or, alternatively,
J
X
j=1
pcj
Cj +
S
X
s=J+1
where Ks is the household consumption of the health input good s, pcj represents
the price of the consumption good j, pkj represents the price of the health input
s, w is the hourly wages. Moreover, Y represents the non-labor income, T p is the
total time available for parents to allocate among work, leisure and child care1
and Y ∗ is the full income of the households as defined by Becker (1981) and also
used by Kassouf (1994).
The solution to the utility maximization problem of the household requires
the assumption of continuous, differentiable and quasi-concave utility functions,
as well as a convex restriction set. It can be shown that under these assumptions
the maximization of the utility function (1) subject to the health production
function (2) and to the full income restriction (4) leads to a unique intrinsic
solution (Kassouf, 1994, p.238).
The first-order conditions associated with this optimization problem can be
used to derive reduced form demand functions for consumption goods C, for
health inputs K, for leisure L, and for health H. The reduced form demand for
children health becomes
H = h(pc , pk , w, Y ∗ , X, ε),
(5)
where ε stands for non-observable attributes.
As discussed earlier, (5) cannot be estimated directly because H, the quantity
of health or health level, is not observable. In this paper, we use anthropometric
measures as a proxy to H, as pioneered by Kassouf (1994) and Kassouf and
Senour (1996). Specifically, we estimate equation (5) using each of three types of
anthropometric scores: weight-for-height, weight-for-age, and height-for-age.
1
Although time spent on children care is part of the time allocation decision, there is no
data available on time spent with children carein Brazil, so this type of time allocation will not
be part of our analysis.
3
3
Data Description
The data set used is the 1996/7 the Living Standard Survey undertaken by the
Brazilian Geographical and Statistical Institute (IBGE). Socio-economic and anthropometric measurements were collected between September 1996 and March
1997, at the household level, for two regions of Brazil: the Northeast and the
Southeast (IBGE, 1998). For this study, we selected from the survey only children of ages 0-12 years who were related to family members in the household.
The anthropometric Z-scores were used as proxies to the health status of the
children. Kassouf (1994) and Kassouf and Senour (1996) pioneered the use of Zscores as proxies to health status when analyzing children from 2-5 years of age,
using 1989 National Health and Nutrition Survey data. She presented evidence
supporting the role of the mother’s education in improving child health in Brazil.
As indicated by Kassouf and Senour (1996), there are three types of anthropometric Z-scores: weight-for-height (WHZ), weight for age (WAZ), and height
for age (HAZ). Each one of these scores are associated with very specific types
of malnutrition, and thus health status. When weight for height (WHZ) is used,
low Z-score values indicate bad health due to acute malnutrition; when weight for
height (WAZ) is used, the health problems are due to medium-run malnutrition;
and when weight for age (HAZ) is used, the health problems are associated to
long-run (chronic) malnutrition, also known as “stunting.”
Low or high Z-scores are the result of comparisons with normal, healthy populations. In the case of a country like Brazil, however, it would be a cause of
concern if the percentage of high Z-scores in the sample were higher than that
found in healthy populations. Z-scores equal to or above −1 are taken to indicate healthy individuals.2 On the other hand, Z-scores below −1 are usually
associated with unhealthy children. In general, the lower a children score, the
worse the health problems he or she is likely to suffer. These health problems can
be classified as severe, moderate, and mild. The children with anthropometric
measures which are from one to two standard deviations below the mean, i.e.
scores between −1 and −2, are likely to have mild health problems. Those with
measures between 2 and 3 standard deviations below the mean, thus with scores
2
In Brazil health problems associated to overweight are not yet a cause of serious concern,
unlike what occurs in developed economies like the USA and European countries.
4
Table 1: Percentage of Children by Z-score Types
Z-Score
ranges
< −3
[−3, −2)
[−2, −1)
> −1
Sample Size
Health
Condition
Severe
Moderate
Mild
Normal
Total
4.7
7.9
19.4
68.1
4687
HAZ
NE
4.5
9.8
22.3
63.4
2686
SE
4.9
5.5
15.4
74.3
2001
Total
1.0
2.4
10.1
86.5
WHZ
NE
1.3
2.4
10.9
85.5
SE
0.5
2.5
9.2
87.9
Total
0.5
4.6
21.3
73.7
WAZ
NE
0.6
6.3
24.7
68.4
SE
0.3
2.3
16.7
80.8
between −2 and −3, are likely to have moderate health problems. Finally, the
children with scores below −3 are likely to have severe health problems.
The Z-scores for the children 0-12 years old are presented in Table 3. The
major problem associated with the Z-scores are the low values for HAZ, which
indicate that 12.6% of the children from the sample suffer from moderate to severe
health problems. The percentage is 14.3% in the Northeast and 10.4% in the
Southeast.3 The data also indicates that the situation does not change if health
condition is measured by WAZ or by WHZ. All three indicators suggest that
stunting – along with its consequences for child health – affects a high percentage
of children in the sample. The situation is particularly severe in the Northeast
while in the Southeast the health level although higher than in the Northeast is
still lower than those of developed economies.4 Stunting is generally associated
with poor economic conditions; its abundance indicates that a large percentage
of children in Brazil – especially in the Northeast – suffer from various infectious
diseases such as diarrhea, cholera and respiratory illnesses.
Table 2 presents the mean and standard deviation of the variables used in the
estimation of the anthropometric demand equations and the estimation of the
Heckman Selection Model used to estimate the household full income.
The variables included in the model explaining the decision of the individuals
in participating in the labor force were: age; age2 ; the number of years of formal
schooling, education; a variable which indicate the position of the individual in the
3
Using 1989 data Kassouf and Senour (1996) found a value of 16% for children 2-5 years old.
Her sample is representative of Brazil as a whole while the sample used in this study covers only
two regions of the country for 1996/7. Alves and Belluzzo (2004) show that infant mortality
rates dropped sharply in Brazil between 1970 and 2000. It could be that the discrepancy
between these two percentages is due to improvements in child health reflected by the rapid
decline in infant mortality rates.
4
For healthy child populations this percentage is 2.5%.
5
Table 2: Description of the Variables
Variables
Mean
s.d.
WAZ
HAZ
WHZ
Southeast
Running Water
Urban
Electricity
Garbage
White
Asian
Parda
Child Sex
Child Age
Log Full Income
WH Father
WH Mother
Mother age < 16
Number of Brothers
Number of children < 12
Mother Education
Private Health Insurance
-0.091
-0.306
0.296
0.427
0.663
0.694
0.876
0.594
0.434
0.002
0.507
0.509
6.209
2.973
7.918
9.178
0.998
2.198
2.700
6.154
0.198
1.427
1.729
1.761
0.495
0.473
0.461
0.329
0.491
0.496
0.044
0.500
0.499
3.784
2.005
4.945
4.929
0.044
1.904
1.904
4.219
0.398
6
household: (Hh for the head of the household, sun for son/daughter, and other
positions within the household are omitted); white to capture the effect of race
(the other races, Asian and parda5 are omitted); southeast, a dummy variable to
distinguish the two regions of the survey: its value is one if the household belongs
to the Southeast region and zero otherwise; urban, a dummy variable indicating if
the household is located in the urban area (value one) or in the rural areas (value
zero); Lnlabor, the logarithm of non-labor income; son 12 indicates the number
of young children ( children less than 12 years old) in the family; and finally We
and WAge are education and age, respectively, of the spouse.
4
Opportunity Cost of Labor and Full Income
Full income, defined by Becker (1965, 1981) as the opportunity cost of available
labor time per month plus any other kind of no-labor income received per month
by the household members, was calculated for each household in the sample. The
hourly wage was multiplied by the number of hours per month available to the
members of the household who are able to participate in the labor force. Some of
the men and women in the sample did not participate in the labor force. Therefore
the opportunity cost of time for those not employed had to be estimated. The
opportunity costs are the wage rates that the individuals not in the labor force
would have received had they been employed. In the wage rate estimation all
men and women between 16 and 70 years old were used.
In order to estimate the wage rate the individuals not in the labor force
would have received we estimated wage equations to men and women separately.
These wage equation were estimated using Heckman’s (1979) two-step procedure
to eliminate the sample selection bias. The first step in the Heckman’s (1976)
procedure the wage equations is to estimate a probit for the participation in the
labor force. Then, the first step coefficients are used to compute a correction
term, which is included as a covariate in the wage equation.
Table 3 presents the estimated selection equation and wage equation for men
and women. Results obtained indicate that participation in the labor force is
negatively related to non-labor income. In the case of women a 1% increase
5
In the Brazilian census, Parda is the general denomination for the various degrees of
white-black racial mixes.
7
Table 3: Coefficient Estimates from Heckman’s Sample Selection Model
Constant
Age
Age2
Education
Household Head
Son or Daughter
Southeast
White
Urban
Wife
Student
Num Children < 12
Log Household Income
Spouse Education
Spouse Age
Lambda
Rho
Sigma
N
Censored Obs.
Selection Equation
Men
Women
0.104
-0.281∗
(0.16)
(0.15)
0.012∗
0.015†
(0.01)
(0.01)
0.000‡
0.000‡
(0.00)
(0.00)
0.024‡
0.043‡
(0.01)
(0.01)
0.239‡
-0.049
(0.09)
(0.07)
-0.125∗ -0.3288‡
(0.07)
(0.07)
0.164‡
0.128‡
(0.04)
(0.04)
0.042
-0.002
(0.04)
(0.04)
0.177‡
0.128‡
(0.04)
(0.04)
0.078
(0.12)
0.0234
0.0442
(0.06)
(0.06)
-0.262‡
-0.176‡
(0.02)
(0.02)
-0.139‡
-0.080‡
(0.02)
(0.02)
0.002
-0.015‡
(0.01)
(0.01)
-0.004‡
-0.001
(0.00)
(0.00)
0.311†
0.777
(0.12)
(0.12)
0.312
0.654
(0.11)
(0.07)
1.000
1.189
(0.03)
(0.05)
5578
6223
3408
3889
8
Wage Equation
Men
Women
-2.851‡
-3.383‡
(0.21)
(0.22)
0.009
0.015∗
(0.01)
(0.01)
0.000
0.000†
(0.00)
(0.00)
0.122‡
0.118‡
(0.01)
(0.01)
0.194†
-0.374‡
(0.08)
(0.09)
-0.0947
-0.4985‡
(0.09)
(0.09)
0.425‡
0.492‡
(0.05)
(0.05)
0.129‡
0.181‡
(0.05)
(0.05)
0.177‡
0.577‡
(0.04)
(0.06)
-0.068
(0.08)
in non-labor income reduces the participation rate by 0.072%. Women in the
Southeast have higher participation in the labor force while married women have
lower participation in the labor force. The significance and sign of num children
< 12 show that having small children in the family is a deterrent to women’s
labor force participation, a result consistent with other studies on the subject,
such as Mroz (1987) and Killingsworth and Heckman (1986), among others. It
is worth to note that some other variables associated with urban-rural dummies
such as urban/rural Southeast and urban/rural Northeast were included in the
model but were not statistically significant. The urban-rural for both regions,
however, was significant. The same occurs with race, which is why only white
was included.
The results for the wage equations for men and women are presented in the
last two columns of Table 3, respectively. In both cases the dependent variable for
the wage equation was the logarithm of hourly wage. The sign and significance
level of the estimated coefficients seem to be consistent with a priori expectations.
Moreover, the LR tests reported in both tables show that the sample selection
would have led to bias in the coefficients had the estimation procedure not been
taken into account.
The wage equation for women shows that almost all variables are statistically
significant and have the expected signs. A Married women earn lower wages than
singles, and white women earn higher wages per hour than black or parda women,
an indication of discrimination in the job market. Education and experience
contribute positively to wage. Men and women in the Southeast earn higher
wages.
Using the coefficient estimates reported in Table 3 we computed the predicted
wage rates for each individual in sample with age between 16 and 70 years. Then,
full income was estimated by multiplying the wage estimate for each member of
the household by 720 (the total number of hours in a month) and plus any nonlabor income. Note that full income is an exogenous variable because it does not
depend on the allocation of time by the household members, while the wage rate
is a given in the labor market.6
6
The procedure reported here is usually used in estimation of wage equation. Kassouf
(1994) and Kassouf and Senour (1996) used this procedure in estimating the role of parents in
children nutrition in Brazil. We use the same arguments she uses to emphasize the exogenous
role of full income in the demand for child health.
9
The estimation of hourly wages and the full income raise some identification
questions in estimating the demand for child health. The strategy used in this
paper differs from the one used by Kassouf and Senour (1996) in two major
aspects. First, in Kassouf’s paper they did not include the number of young
children in the family in the selection equation, which turned out to be highly
significant in our model. Second, Kassouf and Senour (1996) included relation
variables (such as head, wife, daughter and so on) and characteristics of the spouse
both in the selection and in the wage equations, but not in the demand equations.
There Kassouf and Senour argue in favor of using the predicted logarithm of wage
and full income as explanatory variables in the equation for the anthropometric
demand for child health.
In this paper a different identification procedure was used due to the inclusion
of young children in the selection equation. Specifically, we included the mother’s
education as an explanatory variable in the wage equation, and that is why we
chose not to include the estimated wage rate in the demand for health. We
also included convenio as an additional explanatory variable representing private
health insurance coverage of families.7
5
Empirical Results
As discussed earlier, there are types of Z-scores: HAZ (height-for-age), WHZ
(weight-for-height), and WAZ (weight-for-age). Each type of Z-score corresponds
to one dimension of health associated respectively to long run, medium run and
short run, or chronic, malnutrition. As a result, there are three candidates to
proxy the children health level. We estimate a anthropometric demand for child
health for each of these cases.
Table 4 presents the OLS results of the estimation of the reduced form anthropometric demand for child health equations. The t-statistics reported in Table
7
We believe that it is hard to defend the exogeniety of the mother’s education both in
the child health demand equation and in the wage equation; omission of variables related to
ability is a fact in both equations. In both cases the estimated coefficient in the anthropometric
demand equation would be biased and inconsistent. The endogeniety argument also applies to
the convenio variable. The decision to acquire private insurance is part of the decision-making
process of the demand for health and families with a history of child disease will be more eager
to acquire health insurance than others. See Alves and Timmins (2003) for a discussion of this
issue in the context of an analysis of the Brazilian health system.
10
4 were calculated using White’s (1980) heteroskedastic consistent covariance matrix estimator.8 Some aspects of these results are worthy of note. Children living
in the Southeastern region, other factors held constant, have better health than
children living in the Northeastern region.9
Apparently, living in urban areas is not very important to explain the health
level. This explanatory variable, urban, is not significantly different from zero in
the height-for-age and weight-for-age equations. Nonetheless, it is significantly
positive, at the 10% level, in the weight-for-height equation, indicating that it may
have some impact on health. This result is somewhat in line with the findings of
Thomas and Strauss (1992), indicating a positive impact of urbanization on child
height-for-age, but it contrasts with the results of Kassouf and Senour (1996)
indicating a negative impact of urbanization on child health.
Infrastructure showed a mixed impact on child health. Except for running
water, all variables capturing the provision of infrastructure have either positive
significant effect or it is not significant. The results presented in Table 4 indicate
that running water has a positive effect, at the 10% level, in the height-for-age
equation.10 However, the results indicate a negative impact of running water on
weight-for-age and weight-for-height equations, at the 10% and 1% level, respectively.
The variable electricity have a strongly significant positive impact in heightfor-age and weight-for-age, but no significant impact for weight-for-height. Thus,
a household supplied with electricity leads, ceteris paribus, to higher health levels
for children living in it than one that has no electrical supply. The same conclusion
applies to the results for weight-for-age. In the weight-for-height equation the
coefficient of electricity is not significant.
Garbage collection, represented by the variable garbage, has a significant significant positive impact on health according to the weight-for-height and weightfor-age equations. In the height-for-age case, however, its effect was not significant. Thus, with the exception of the unexpected result for the running water
8
OLS with constant variance did not seem to be supported by the sample data. Standard
errors changed substantially when White robust standard errors were estimated.
9
This result is consistent with the results presented in Alves and Belluzzo (2004), which
indicate that the Northeast is well behind other regions of the country in terms of child health.
10
This result corroborates, to some extent, the findings of Alves and Belluzzo (2004) indicating a positive impact of running treated water on infant mortality reduction.
11
Table 4: Anthropometric Health Demand Equations
Constant
Southeast
urban
running water
electricity
garbage
White
Asian
parda
Sex
child age
log full income
WH father
WH mother
num siblings
mother age < 16
mother education
R-squared
F-statistic
Sample Size
HAZ
-0.360
(-0.52)
0.014
(0.25)
-0.171†
(-2.05)
0.152∗
(1.81)
0.546‡
(6.70)
0.008
(0.11)
-0.205
(-1.56)
0.595
(0.77)
-0.126
(-0.98)
-0.158‡
(-3.15)
-0.007
(-0.87)
0.293‡
(5.46)
0.008
(1.48)
0.024‡
(4.81)
-0.054‡
(-3.51)
-1.824‡
(-2.82)
0.020†
(2.25)
0.102
28.89
3980
12
WAZ
-0.568
(-1.21)
0.126‡
( 2.76)
-0.001
(-0.01)
-0.098∗
(-1.61)
0.267‡
(4.20)
0.247‡
(4.21)
0.195†
(1.96)
0.320
(1.04)
0.118
(1.22)
-0.143‡
(-3.59)
-0.041‡
(-7.32)
0.143‡
(3.44)
0.025‡
(5.71)
0.041‡
(9.16)
-0.092‡
(-8.14)
-0.893†
(-2.05)
0.022‡
(3.18)
0.176
52.91
3976
WHZ
-0.03
(-0.05)
0.209‡
(2.94)
0.168∗
(1.65)
-0.313‡
(-3.26)
-0.119
(-1.39)
0.376‡
(4.21)
0.400†
(2.11)
-1.56†
(-2.52)
0.203
(1.09)
-0.136†
(-2.24)
-0.031‡
(-2.94)
-0.084
(-1.38)
0.032‡
(4.61)
0.028‡
(4.19)
-0.074‡
(-4.53)
0.077
(0.14)
0.022†
(2.04)
0.06
11.94
3022
variable, our results indicate a positive impact of the infrastructure variables on
child health. Better infrastructure, other factors held constant, improves child
health.
The results presented in Table 4 indicate that race might not be a very important factor explaining health differences. There is some indication that white
children have somewhat better health levels than black, pardos and asian children
when weight-for-height and weight-for-age are used as proxies. Asian children
seem to do worse than other races when weight-for-height is used, but there is no
significant differences when the other two proxies are used. All of the remaining
race coefficients are not statistically different from zero. In contrast, gender does
seem to significantly impact health: all three proxies for health are negatively
related to gender, indicating that girls are healthier than boys. Interestingly, this
result contrasts with the findings of Kassouf and Senour (1996) indicating that
there is no differences in health due to gender.11
The variable child age has negative sign in all three equations, even though
it is not significant in the height-for-age equation. The coefficient of logarithm
of per capita full income, indicated by log full income in Table 4, is positive and
highly significant in the height-for-age and in the weight-for-age equations. In
the weight-for-height equation it was not significant. These results indicate that
stunting and wasting are positively associated with poverty. Higher per capita
income leads to better health, as predicted by the reduced form, equation (5) of
the theoretical model. To the extent that weight-for-height captures the long-run
impact of malnutrition, the results reported for this variable make a lot of sense.
Poverty takes a long-term toll in terms of child heath; low income per person in
the household leads to lower health status even in the long run.
In all of the equations both the weight-for-height of the parents are important factors in explaining the health differences. They have positive effects on
health, except only for the weight-for-height of the father in the height-for-age
equation. Even though this finding suggests that genetics plays a role in differences in health among children, it is important to note that it is very likely
that long-run overall poverty of parents manifests itself in the weight-for-height
Z-scores. These variables have been used as proxies for unobserved family back11
See Kassouf and Senour (1996, p.824).
13
ground characteristics.12 Controlling for family characteristics is important in the
analysis of the estimation of the demand for child health. Behrman and Wolfe
(1984) and Behrman (1990), for instance, argue that the impact of education may
be overestimated because of unobserved endowment and family background.
Low-income families of developing countries are usually large. Indeed, in our
sample the size of the family is larger for low-income households than for highincome households. Parents of low-income families have to take care of and feed
a larger number of children than parents of high-income families.13 Variable
num siblings is the number of brothers and sisters of the corresponding child
observation in the regression equations. The sign and the significance level of the
coefficients of this variable indicate that larger numbers of children contribute to
lower child health levels. This result clearly indicates that an additional child in
the family will lead to lower health levels for all of the children in the family.
The variable mother age < 16, indicating if the mother is younger than 16
years, also has a detrimental effect on child health when health is measured by
height-for-age and weight-for-age. When health is measured by the weight-forheight, however, it has no significant impact on child health. Younger mothers
are less experienced in dealing with children, leading to a lower health for their
children.14
The variable mother education has a positive and significant coefficient in all
of the demand equations. This result is quite strong, as would be expected on the
basis of several other studies indicating the positive impact of the mother’s education on the improvement of child health in developing countries. See, for example,
Behrman and Wolfe (1984), Thomas, Strauss, and Henriques (1990), Adelman
and Garcia (1993), Behrman and Deolalikar (1988), Kassouf and Senour (1996)
and Kassouf (1994). The inclusion of full per capita income and mother’s education and full income might look contradictory, because the mother’s education
enters as an explanatory variable in the mother’s wage equation while the predicted mother’s wage enters into the composition of household full income. The
inclusion of the mother’s education in the health demand equations is an attempt
12
See Thomas and Strauss (1992, p.317), and Kassouf and Senour (1996, p.825).
In the sample the 10% highest income bracket has 2.3 children per household, while the
10% lowest has 2.6.
14
Horton (1988) found a positive relation between mother’s age and child health while Strauss
(1990) did not find any relationship.
13
14
to capture the net contribution of the mother’s education to child health. In the
full income variable the contribution of the mother’s education explains child
health indirectly. Therefore, the mother’s education would have both a direct
and an indirect dimension, the latter of which derives from the value of education
in the job market. In our study, however, the results indicate that the direct
effect of the mother’s education is not so strong, so that the indirect effect might
be stronger than would be expected.
The number of brothers and sisters, num siblings, is detrimental to child
health. In all Z-scores specifications the coefficient for num siblings is negative and
statistically significant. This result indicates that families with a large number
of children, others factors held constant, tend to have less healthy children. The
gender variable, sex, indicates that boys are in worse health status than girls in
all demand equations; while in all of the demand equations the variable age has a
negative and significant coefficient, indicating that as a child gets older its health
situation tends to worsen .
6
Conclusion
The study clearly shows that child health differs quite substantially if he/she lives
in the Northeast or in the Southeast. Regional effects play a strong role in child
health. History and culture are fundamental forces explaining child health differentials. Among the socioeconomic variables education, in general, and mother’s
education in particular, play a strong role in explaining child health. Mother’s
education plays a direct and an indirect role in the explanation of child health;
there is a direct effect of mother’s education through the health care of the children. Furthermore, higher education leads to higher per capita full income which
allows expanding health of the children.
Policy actions leading to improve education of the population and policies
leading to the permanence of the young women for additional years in the school
guarantying improved education are important contributors to child health. The
analysis has shown that pregnancy of young women is an important factor explaining lower child health. Thus, if the youngest and specially the women are
incentivated to remain additional years studying they not only will get more education as well as be able to prevent unwanted pregnancy of the youngest women
15
a major factor explaining low child health.
A strong point against large family is made by the results large number of
brothers and sisters are strongly detrimental to child health. The key is education
and education of the women creating an incentive structure to keep them more
years in school. Present day policy incentive structure is biased towards larger
families.
Government programs provide incentive to large family, by having instruments
for income transfer in the social program based on number of children. Educational programs distribute income for poor families to maintain their children in
primary school and the income distribution scheme rewards the same amount of
money per children independently of the number of children per family, providing
incentives for larger families.
Brazilian social programs should be unified in their goals. Furthermore they
should change the income redistribution incentives in order to prevent large families incentive and to improve education and incentives in such a way as to improve
the most the schooling of young women.
References
Adelman, H., & Garcia, M. (1993). Poverty, household food security, and nutrition in rural pasquitan (Research Report No. 96). Washington, DC:
Internation Food Policy Research Institute.
Alves, D., & Belluzzo, W. (2004). Child health and infant mortality in brazil.
Final Seminar on Child Health, Poverty and the Role of Social Policies.
IDB. Washington, DC.
Alves, D., & Timmins, C. (2003). Social exclusion and the two-tiered health care
system in brazil. In J. Behrman, A. Gaviria, & M. Szekely (Eds.), Who is
in and who is out. Washington, DC.: IDB.
Becker, G. (1965). A theory of allocation of time. Economic Journal, 75, 493-517.
Becker, G. (1981). A treatise of the family. Cambridge: Harvard University
Press.
16
Behrman, J. (1990). Women schooling and non-market productivity: A survey
and reapprisal. mimeo, World Bank. Washington, DC.
Behrman, J., & Deolalikar, A. (1988). Health and nutrition. In H.Chenery &
T. N. Srinivasan (Eds.), Handbook of development economics. New York:
North Holland.
Behrman, J., & Wolfe, B. (1984). More evidence on nutricional demand: Income seems overrated and women’s shooling underemphasized. Journal of
Development Economics, 14, 105-128.
Heckman, J. (1979). Sample selection bias as a specification error. Econometrica,
47 (1), 153-162.
Horton, S. (1988). Birth order and child nutritional status: Evidence from the
philippines. Economic Development and Cultural Change, 36, 341-354.
IBGE. (1998). Pesquisa de padrões de vida. Rio de Janeiro.
Kassouf, A. (1994). A demanda de saúde infantil no brasil por região e setor.
Pesquisa e Planejamento Econômico, 24 (2), 235-260.
Kassouf, A., & Senour, B. (1996). Direct and indirect effect of parental education
on malnutrition among children in brazil: A full income approach. Economic
Development and Cultural Change, X (X), 817-838.
Killingsworth, M. R., & Heckman, J. (1986). Female labor supply: A survey.
In O. Ashenfelter & R. Layard (Eds.), Handbook of labor economics (p.
103-204). New York: Elsevier Science Publishers.
Mroz, T. A. (1987). The sensivity of an empirical model of married women’s
hours of work to economic and statistical assumptions. Econometrica, 55,
765-799.
Thomas, D., & Strauss, J. (1992). Prices, infraestructure, household characteristics and child age. Journal of Development, 39, 301-331.
Thomas, D., Strauss, J., & Henriques, M. H. (1990). Child survival, height for
age and household characteristics and child height. Journal of Development
Economics, 33, 197-234.
17
White, H. (1980). A heteroskedastic consistent covariance matrix estimator and
direct test of heteroskedasticity. Econometrica, 48, 817-838.
18