DOES MOTHER’S EDUCATION MATTER IN CHILD’S HEALTH? EVIDENCE FROM SOUTH AFRICA doc

DOES MOTHER’S EDUCATION MATTER IN CHILD’S
HEALTH? EVIDENCE FROM SOUTH AFRICA
1
patricia medrano*, catherine rodri´guez
†
and edgar villa
‡
Abstract
Using the 1993 South Africa Integrated Household Survey, this paper studies the effect that
mother’s education through the knowledge channel has on children’s health using height for age
Z-scores as health measure. Under a two-stage least square methodology we find that an increase
in 4 years on mother’s education (approximately 1 standard deviation) will lead to an increase of
0.6 standard deviations on her child’s height for age Z-score. We also find, as the medical literature
suggests, support for the hypothesis that mother’s education is more important for children older
than 24 months of age.
JEL Classification: I12, I21, O15, D1
Keywords: Education, health, Z-score
1. INTRODUCTION
Poverty can be thought of as a vicious circle. Low-income households have less education
and are probably less healthy than wealthier ones, making it very difficult for them to leave
that poverty state through their own work and effort. It has been previously established
that investments in children’s health and education could help break this circle by
enhancing individual’s future income through their future productivity. For instance,
Currie and Hyson (1999), Case et al. (2003) and Behrman and Rosenzweig (2004) have
found strong links between child’s health and future income, education attainment and
adult health. Not surprisingly, many of the Millennium Development Goals are related to
the improvement of children’s education and health.
2
In this paper we touch on this
important subject and asses the impact that investment in mother’s education through the
knowledge channel has on children’s health measured as height for age.

Conditions within a household are important determinants of a child’s health and as
one might expect parents play a central role in “producing children with good health”.
Specifically, the “mother” has been described as the most important health worker in the
household. Nonetheless, how well a mother performs this task may depend on factors
1
We are grateful for the advice and helpful comments of Julie Anderson Schaffner and the
participants of the Development Seminar at Boston University on an earlier draft. We also thank
the comments and suggestions received from an anonymous referee. All views and errors are
exclusively those of the authors. The authors also acknowledge the financial support from the
Iniciativa Científica Milenio to the Centro de Microdatos (Project P07S-023-F).
* Centro de Microdatos, Department of Economics, University of Chile; pmedrano@
econ.uchile.cl
†
Universidad de los Andes; corresponding author:
‡
Pontificia Universidad Javeriana;
2
Among the goals we find the reduction of child mortality, eradicate hunger, achieve universal
primary education and improve maternal health.
South African Journal of Economics Vol. 76:4 December 2008
© 2008 The Authors.
Journal compilation © 2008 Economic Society of South Africa. Published by Blackwell Publishing Ltd, 9600 Garsington
Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.
612
such as her own health, the resources that are available to her and her schooling. In
particular, it has sometimes been argued that schooling equips the mother with a specific
knowledge that enhances her ability to generate healthier children. This paper quantifies
the relationship between this key factor and child’s health using South African data from
the 1993 Integrated Household Survey (LSMS).
The magnitude of this relationship has important policy implications. If there is a

positive effect of mother’s education as knowledge on children’s health, certain type of
policies concerning the diffusion of specific information at community levels can be
called for. For example, one important channel through which mother’s education as
knowledge can affect child’s health at a community level is through the usage of health
facilities which can serve as a complement or even as a substitute for the mother’s
education. Therefore, a policy concern that we examine is the impact of health facilities
interacted with mother’s education as knowledge. Finally, motivated by the U shape
pattern in children’s health that the medical and public health literature suggests exists,
we study whether mother’s education affects in different ways children of different age
groups.
Even though this question has been previously addressed in the literature, the South
Africa of the early 90s is an interesting place to re-examine it because of the wide disparity
in income and health between poor and rich households. During the Apartheid era,
South Africa was an environment where the Black communities in the rural areas
did poorly on most of the indicators of well-being, one of them being health status.
The results found show that mother’s education has a positive and significant impact
on children’s long-run health proxied by their height for age Z-score. We find for an
ordinary least squares (OLS) regression that mother’s with 4 more years of education
(approximately 1 standard deviation) increase on average 0.1 standard deviations a child’s
height for age Z-score. This effect increases to 0.6 standard deviations when mother’s
education, household disposable income and father’s education are instrumented
adequately. We find that this effect is maintained for children between 24 and 72 months
of age while vanishing for younger children.
Following this introduction we review some of the existing literature on this and
related topics. We then proceed to develop a conceptual framework to explain the possible
transmission channels of mother’s education and the estimation problems that should be
acknowledged in the empirical specifications. Section four describes the data set we use
for our empirical application and section five presents our findings. Section six ends with
conclusions.
2. LITERATURE REVIEW

The empirical literature on human capital investment generally focuses on education and
health investments using reduced form models that integrate the health production
process with a model of household choice. The demand studies for health outcomes
usually evaluate the impact that household and community level characteristics, like the
usage and availability of health facilities, may have on distinct health measures. Among
these, many have focused on the effect that mothers’ education has on children’s health
measures such as birth weight, height and weight for age Z-scores or nutrient intake. A
comprehensive review of earlier studies can be found in Behrman and Deolalikar (1988)
and Strauss and Thomas (1995). In this section, however, we review some studies that use
613South African Journal of Economics Vol. 76:4 December 2008
© 2008 The Authors.
Journal compilation © Economic Society of South Africa 2008.
height for age Z-scores as a proxy of children’s health and hence are closely related to the
present paper.
Probably, the closest related paper in the literature to this study is Glewwe (1999).
Using cross-sectional data from Morocco, he estimates the impact that mothers’ health
knowledge has on children’s long-term health status proxied by height for age Z-scores.
The empirical approach taken by the author is careful in two distinct aspects. First, the
database provides information that can help the researcher distinguish between general
numeracy and literacy skills of the mother and their specific knowledge on health issues.
Second, given that variables such as health knowledge, household income and literacy and
numeracy skills may be endogenous, the author uses an instrumental variable approach to
correct for such possibility.
3
While simple OLS coefficients suggest that the knowledge
channel is not important in determining child’s health; when endogeneity issues are
addressed the IV coefficient on mother’s knowledge increases and is highly significant.
In another study, Barrera (1990) assesses the efficiency and allocative effects of
maternal education. Using data from the Philippines he finds a positive impact of
maternal education, which is stronger for younger children. Similarly, Thomas et al.

(1991) study the impact of mothers’ education in child’s height for age Z-score. The
authors argue there are three channels through which mother’s education can improve a
child’s health: income augmenting effects, information processing effects and interactive
effects with community services. Using information on Brazilian children they find that
almost all the impact of maternal education can be explained by indicators of access to
information which are proxied by the availability of media services in the community.
Using information on children from Côte d’Ivoire, Strauss (1990) evaluates whether
differences in height and weight for age Z-scores can be explained by differences in the
level of education of the parents, local wages and the availability in health, schooling and
water infrastructure. As the previously cited studies, the author worries about the
assumption of exogeneity of observable control variables. In order to reduce any potential
bias to the coefficient of parents’ education caused by such endogeneity problems, Strauss
(1990) includes mother’s standardised height to capture genetic and background
characteristics of the family. Moreover, he also presents fixed and random effects
estimations in order to test for the sources of correlation. The results show that after
controlling for mother’s height, her education has a positive but small effect on child’s
height.
Contrary to all these studies, Wolfe and Behrman (1987) do not find an important
effect of mother’s education on a child’s height for age Z-score. Using a special sister
sample from Nicaragua they can control for the mother’s background variables and find
that the significance and magnitude of the impact of education disappears. They conclude
then that mothers’ education is simply serving as a proxy of her background and only has
a significant effect on calorie intake measures.
In a novel and recent study, Chen and Li (2006) estimate the nurturing effect of
mothers’ education on child’s health again proxied by height for age Z-scores. According
to the authors, while the nurturing effect should be related to mother’s education, the
nature effect is caused by selection or omitted variables. In order to obtain the former
effect, the authors use information on Chinese adopted children which are of course
3
Among the different instruments used by the author one finds household assets, exposure to

mass media and the education of siblings of the mother among others.
South African Journal of Economics Vol. 76:4 December 2008614
© 2008 The Authors.
Journal compilation © Economic Society of South Africa 2008.
genetically unrelated to the nurturing parents implying that endogeneity problems should
no longer exist. Using simple OLS regressions the authors find a positive and significant
nurturing effect of the mother. Moreover, comparing the estimates with a sample of
biological children they conclude that most of the impact on children’s health is not given
by genetic factors but through the education channel.
This literature has also addressed the question of gender bias. Thomas (1994) using data
from United States, Brazil and Ghana finds that maternal education and non-labour
income have a bigger impact on the height of a daughter relative to a son while paternal
education has the opposite effect. Namely father’s education has a bigger impact on a son
relative to a daughter. Duflo (2000) gives evidence that households do not function as a
unitary entity. Using data from South Africa she finds a positive effect on girl’s
anthropometric measures if a grandmother, rather than a grandfather, receives cash
transfers by a social security program while no apparent effect is found for boy’s health
status.
The general evidence found in the literature allows us to argue in favour of the
hypothesis that there exists a positive effect of mother’s education on height for age
measures of children. The biggest concern is probably that reduced form estimates may
understate the impact of this variable given its probable correlation with a child’s genetic
background. The following sections will address these concerns and evaluate whether the
positive effects are also found in a representative sample of South African children which,
to the best of our knowledge, has not been carried out yet.
3. CONCEPTUAL FRAMEWORK
As mentioned above, in this paper we focus on the role that mother’s knowledge plays in
enhancing future household productivity through child health investments. However,
we should be cautious in interpreting the different channels through which mother’s
education could affect children’s health. It is generally accepted that mother’s education

can affect child health through the following five channels:
(i) It increases economic resources of the family by increasing own earnings.
(ii) It increases efficiency in the usage of available health facilities.
(iii) It can affect household preferences.
(iv) It improves allocation of resources due to better knowledge and information.
(v) It can indicate wealth status and assortative mating.
Disentangling these different channels is crucial to assess the specific impact of mother’s
education as knowledge on children’s health (channel iv).
Health Production and Household Model
In order to study the determinants of children’s health, the literature has traditionally
used a standard household utility maximisation approach that depends on child’s health,
which in turn is determined by a health production function.
4
Following this literature,
we assume that household i’s utility is an increasing function of the consumption of child
health (H
i
) given other non-health-related commodities of the household (Z
i
) and
observable household characteristics C
i
. That is, utility will be described by the following
4
A detailed description of these models can be found in Strauss and Thomas (1995) and Thomas
et al. (1991).
615South African Journal of Economics Vol. 76:4 December 2008
© 2008 The Authors.
Journal compilation © Economic Society of South Africa 2008.
standard function U

i
= U(H
i
,Z
i
;C
i
). As mentioned, child’s health is the output of a
production function that depends on a child’s biological endowment (e), the effective
units of mothers input (X*) and health services at the community level (HF). We
summarise it as:
H = h( , X *, HF )
iiii
ε
(1)
Reasonably we assume that greater biological endowment as well as effective units of
mother’s education and usage of health facilities infrastructure increase a child’s health,
i.e. Dh/De > 0, Dh/DX* > 0 and Dh/DHF > 0.
Moreover, we assume that the effective units of mothers input depend on two factors:
household goods such as appropriate food that serve as inputs for child care (W) and
household child care knowledge (CK) which uses all these inputs in an efficient manner.
Under this assumption we can express X
i
* = k(W
i
,CK
i
) where k(·) represents the
functional form by which the effective units of mothers input is formed. Crucially we
assume that household child care knowledge is determined by formal education of

household members. This can be justified in the sense that skills to read and write within
a household generate better child care knowledge and therefore a more efficient
technology. This can be formalised as
CK = r(MED , FED )
iii
(2)
where years of mother’s and father’s formal education are denoted by MED and FED,
respectively. We conjecture that greater formal education increases household child care
knowledge, i.e. Dr/DMED > 0 and Dr/DFED > 0. Hence, a functional relation can be
represented by:
X
*
= k(W , r(MED , FED ))
ii ii
(3)
We conjecture about the following relations that affect effective units of mother’s input:
(i) Inputs W should have a marginal positive effect on effective units of mother’s input.
(ii) Mother’s and father’s formal education should have a positive effect on effective
units of mother’s input only through child care knowledge, i.e. DX
i
*/DMED > 0 and
DX
i
*/DFED > 0.
To close the model, we assume a traditional budget constraint for the household of the
form: Y
i
= P
W
W

i
+ P
Z
Z
i
+ P
HF
HF
i
; where P
j
is the vector price of goods j = W, Z, HF and
Y
i
is disposable income of household i. The household will naturally maximise its utility
subject to the production function of health (1) and its budget constraint. The model
thus yields a reduced form demand function for child health given by:
H = g(Y , P , P , P ; HF , MED , FED , C , )
iiWZHFi iiii
ε
(4)
where recall that e
i
represents the unobservable biological health endowment of a child
living in household i. Notice that HF usage increases directly the overall utility of the
household as it increases child health but can have an ambiguous effect on overall utility
since households have to pay for using these facilities as assumed in (4).
South African Journal of Economics Vol. 76:4 December 2008616
© 2008 The Authors.
Journal compilation © Economic Society of South Africa 2008.

Empirical Specifications and Methodology
A linear specification of the demand function for child health is given by the following
equation:
H = + MED + Y + C + +
i01 i1i i2 i
ββ δ ε
ddP
3
(5)
where prices are lumped together as a row vector P and C
i
is a vector of observable child
and household characteristics such as father’s education; usages of health facilities; gender
of child, race and age of child; age of mother; if head is a widow mother; if father of child
is absent; number of siblings and the quality of the house they live in captured by number
of rooms, household usage of electricity, usage of tap water and availability of a toilet. The
parameter of interest is b
1
which we conjecture to be non-negative. Importantly, we
interpret b
1
as a marginal causal effect.
For a given cross-section of households equation (5) may be estimated by OLS.
Nonetheless, several parameters would be inconsistently estimated due to endogeneity
problems. However, notice that disposable income Y
i
can be endogenous since it can be
correlated with e
i
for various reasons. One such reason is that households that have had

greater wealth in the past tend to have greater disposable income currently. This greater
wealth in the past could have generated healthier household members, in particular the
mothers, which in turn inherit these health endowments to their children when born
generating a correlation between Y
i
and e
i
. MED
i
and FED
i
can also be correlated indirectly
with e
i
since both are surely positively correlated with disposable income Y
i
. Moreover,
MED and FED can be directly correlated with e
i
since formal education can be correlated
with mothers health, which in turn is correlated with child’s health. Finally, HF
i
usage can
be negatively correlated with e
i
since health facilities are required more for unhealthier
children.The vector P can be correlated with e
i
through disposable income and HF. Notice
that omitting P and HF in (5) still allows us to consistently estimate b

1
if we are willing to
assume (something that is reasonable) that HF and P are only correlated with e
i
through Y
i
and not directly correlated with MED and FED.
To solve these endogeneity problems we propose an instrumental variables approach
where the instruments for MED and FED are: (i) exposure of the household to mass
media; (ii) expenditure in newspapers and entertainment; (iii) if the main language in the
household is Afrikaan or English; and (iv) whether someone in the household has moved
or migrated for academic reasons.
5
The first two instruments have been previously used
and proven useful in the literature by authors such as Glewwe (1999) and Thomas et al.
(1991). The third instrument is a valid one in the South African context given its
particular history. As expressed by Keswell (2004) one of the biggest effects of Apartheid
was the extreme economic inequality generated between race groups. Such inequality is
clearly observed in the differences between average educational attainments across races.
For instance, using information from 1995 Lam (1999) shows that while the average
years of education for non-Whites was 7.22, for Whites it was 11.63. Given that language
is closely related with race, it is expected that its correlation with education should be
high. The fourth instrument could be correlated with the household preference for
education and with the household disposable income which is also instrumented.
5
It would have been ideal to use as additional instruments variables such as mother’s or father’s
siblings’ education. Unfortunately such information is not available to us.
617South African Journal of Economics Vol. 76:4 December 2008
© 2008 The Authors.
Journal compilation © Economic Society of South Africa 2008.

Moreover, we propose ownership of durable goods by the household as instruments for
disposable income (Y
i
). Specifically, we use the following variables: if the household owns
the house in which they live in and the number of motor vehicles, refrigerators, stoves,
telephones and television sets that they report to own. This approach to instrument
disposable household income is also taken in Glewwe (1999).
All instruments can be criticised depending on what one is willing to assume. For
example, a possible critique to our language instrument for MED or FED is that language
of the White minority can also be correlated with accumulated wealth and correlated with
disposable income which in turn is correlated with a child’s endowment. Moreover, one
can believe that household language can be correlated with a child’s biological health
endowment through other channels like a child’s race and gender. Nonetheless, recall that
a child’s race and gender are control variables included in (5) and therefore this potential
pitfall is avoided. Given that more instruments are available than potential endogenous
variables, standard tests of over-identification will be provided in the Results section in
order to empirically try to validate their usage.
4. DATA DESCRIPTION
Although South Africa has better human development indicators than other African
countries such as Kenya and Nigeria, its people still lack basic services and there is much
work needed in order to achieve better development results. Some basic comparative
indicators for 2005 can be seen in Table 1 below. As observed, while clearly South Africa
is more developed than its Sub-Saharan peers, when compared to other developing
countries such as those from Latin America the reality is the opposite. For instance, life
expectancy is only 48 years while the average in Latin America reaches 72. Related to
children’s health the situation is not different either. Specifically, child’s health still is far
behind acceptable standards. While infant mortality of children under 5 years of age in
South Africa is 68 per 1,000 live births, this statistic in Latin America and high-income
countries is 26 and 5.8, respectively. Moreover, according to information in the World
Development Indicators, child mortality in South Africa has increased in the past years.

6
These simple statistics imply that the improvement of children’s health in the country
is imperative. One such way is perhaps through the increase in mothers’ education if it is
proven to be an important variable. To answer this question, the data used in this paper
6
Specifically, the estimates suggest that infant mortality and mortality under 5 in the year 2000
was 50 and 63 per 1,000, respectively.
Table 1. Comparative human development indicators, 2005
South
Africa
Sub-Sahara
Africa
Latin
America
High-income
countries
Life expectancy at birth (years) 48 47 72.5 79
Fertility rate (total births per woman) 2.5 2.3 1.3 0.7
Mortality rate, infant (per 1,000 live births) 2.8 5.3 2.4 1.7
Mortality rate, under 5 (per 1,000 live births) 55 96.4 26.2 5.8
Infant mortality (per 1,000 live births) 68 163 30.8 6.9
Primary completion rate 89 60.8 98.5 97.4
School enrollment/secondary education (gross) 84.9 31.7 87.6 101.4
School enrollment/tertiary education (gross) 14.4 5.1 29.3 66.8
Source: World Bank, World Development Indicators.
South African Journal of Economics Vol. 76:4 December 2008618
© 2008 The Authors.
Journal compilation © Economic Society of South Africa 2008.
comes from the 1993 South Africa LSMS. This is a nationally representative, multi-
purpose household survey conducted by the World Bank which randomly selected 9,000

households from all races and areas throughout the country. It contains data on several
subjects including socioeconomic variables for all members of the households,
anthropometric measures for all children aged less than 7 years as well as community level
characteristics.
Although the LSMS data set contained weight and height measures for each child, our
study is not based on these simple measures. There are several reasons for this decision.
First, the weight of a child relative to the age is a short-term measure of child nutrition
since it responds to changes in health and nutritional status very quickly. Second, the
effect of health status on height is cumulative and can be perceived only in the long run
making it difficult to study the effect for younger children. Since none of these two
anthropometric measures were adequate, we decided to base our measure of child’s health
on one that would allow us to assess the relative poverty in South African children’s health
compared to a standard well-nourished population such as that of the children in the
United States. Hence, we used the height for age Z-scores based on the information
provided by the National Center for Health Statistics from the United States. This
measure reports the standardised deviation of the height of the child with respect to the
median height of the age and gender group in a reference population of well-nourished
children such as that of the United States and is computed in the following manner:
H
height mean
std.dev
i
i
g
g
=
−
(6)
where i indexes each child in the survey and g indicates the age and gender group that
he/she is part of. This measure is commonly used in the literature and has the advantage

that provides useful information of both short- and long-term health status (Thomas
et al., 1991).
7
Under this specification a Z-score of zero implies that the child’s height is equal to the
median height of a well-nourished child; while a negative score would imply that the
child has a poorer health compared to that of the median health of a well-nourished
population. Following the recommendation of the World Health Organization we
focused on children between 6 months and 6 years of age. Limiting the analysis of height
measures in this interval is due to measurement errors for the new born and
environmental factors that affect the older ones. We also dropped from the study children
in the highest and lowest 1% of the Z-scores since their values seemed very unrealistic and
were probably due to measurement errors.
8
After all these recommendations were taken,
7
In principle, another alternative measure that can be used to answer the question of interest is
the analogous weight for age Z-score. However, it is normally catalogued as a short-run measure
that heavily depends on family’s current income. It is expected then that the effect of mothers’
education on this variables should arise through the income channel. This would imply that
regressions that control for this channel should yield an insignificant coefficient on mothers’
education. We carried out these regressions and find that this is in fact the case for South Africa.
They are available upon request.
8
This last restriction in the data used made us loose a total of 73 observations. Children with
Z-scores values such as -112.28 were dropped. It is simply unrealistic to believe a South African
child is 112.28 standard deviations behind a well-nourished American child. However, as
619South African Journal of Economics Vol. 76:4 December 2008
© 2008 The Authors.
Journal compilation © Economic Society of South Africa 2008.
we were left with the Z-scores of 3,912 children between the ages of 6 months and 6 years

that lived in 1,729 different households.
As previously mentioned, the most important aspect that characterised the recent
history of South Africa is the legacy of Apartheid. It was a discriminatory system
institutionalised in the late 1940s that significantly hindered the development of the
country and specially that of its Black population. Race laws touched every aspect of
social life and the population was mainly classified into one of three categories: White,
Black (African) or Coloured (of mixed decent). Two important aspects influenced by such
policy were both education and health of their citizens.
Indeed, descriptive statistics in Table 2 show us that South African children in 1993
had a poorer health status compared to that of American children.
Among the four races that can be distinguished in the survey, the difference with a
well-nourished child is substantially smaller for White children while Black children have
the lowest Z-scores of the four races. Dividing the sample according to the region in which
the household resided shows that children from rural regions appeared to be better-off than
those who live in urban areas. Finally, it is worth mentioning that in this society girls seem
to be relative better off than boys in regards to their height for age Z-scores measures.
As mentioned before our prime interest is the effect that mother’s education has on
children’s health. Table 2 also shows that the mean education for the women in the
sample is around 7 years. This average is considerably different among races; while White
women attain on average 11 years of education, Black women only attain 6 years. This is
an expected result given that during the Apartheid era Black women were discriminated
by their race, class and gender. There are also some differences between the average
robustness check all estimations were carried out maintaining these children and the main
conclusions remain intact. Results are available upon request.
Table 2. Height for age Z-scores and mothers’ average
education
Number of
children
Mean child
height Z-score

Mother’s average years of
education
All children 3,912 -1.154 7.405
(1.976) (3.966)
By race
Blacks 3,220 -1.291 6.917
(1.944) (3.859)
Coloured 317 -1.114 8.249
(1.784) (2.779)
Indian 100 -0.283 10.212
(2.091) (2.757)
Whites 275 0.090 11.124
(2.006) (4.183)
By location
Rural 2,325 -1.387 6.465
(1.953) (3.877)
Urban 737 -0.990 8.388
(1.912) (3.683)
Metro 850 -0.657 9.124
(1.990) (3.652)
By gender
Female 1,907 -1.056 7.395
(1.986) (3.911)
Male 2,005 -1.247 7.415
(1.962) (4.018)
Standard deviations in parenthesis.
Source: 1993 South Africa Integrated Household Survey.
South African Journal of Economics Vol. 76:4 December 2008620
© 2008 The Authors.
Journal compilation © Economic Society of South Africa 2008.

education levels of the mothers depending on the area of residence. Mothers of children
living in urban areas attain on average higher education levels.
Other socioeconomic variables that will be used in the empirical study are displayed in
Table 3. From the sample used in the study one can observe that almost 40% of children
have an absent father and that fathers’ average education is approximately 7.9 years.
9
Looking at other characteristics of the household we observe that total per capita
monthly consumption is 95.45 Rands,
10
only 43% of the households drink water from
the pipe or from water vendors and 35% has either a toilet or an improved latrine as their
sanitation system inside the house. Access to hospitals in 1993 was very limited since only
7% of the households had access to them while 37% had access to other health facilities
such as Maternity Clinics and Pharmacies. As for the descriptive statistics of the
instruments used, it can be observed that the monthly expenditure in mass media was
13.7 with a high standard deviation.
11
The ownership of durable goods are highly related
with current income such as number of motor vehicles, while number of television sets or
number of fridges is generally low with the exception of number of radios.
5. EMPIRICAL RESULTS FOR MOTHER’S EDUCATION AS KNOWLEDGE
OLS Results
As mentioned above our primary interest is to understand the reduced form impact that
mother’s education (through the knowledge channel) has on children’s health between 6
9
Following Glewwe (1999) and in order to avoid loosing the high proportion of children without
fathers, which could lead to important selection bias, the average population values of education
were replaced instead of the missing and a dummy variable indicating that the father was absent
was included in all regressions.
10

This amounts to almost 237 Rands of 2008 or US$30.
11
This amounts to almost 39 Rands of 2008 or US$4.13.
Table 3. Socioeconomic statistics of the households
Mean Standard deviation
Child’s height (24 months) 82.14 6.105
Child’s height (36 months) 90.75 5.837
Child’s height (72 months) 106.87 6.715
Child’s age (months) 36.86 18.363
Child’s weight 13.16 4.127
Mother’s age 30.15 7.495
Father absent 39.32
Father’s education 7.29 4.531
Father’s age 37.81 9.017
Total number of children in the household 1.83 1.012
Total pc consumption 95.64 73.944
Ownership of house they live in 76.41% –
Drinks tapwater 42.31% –
Proper sanitation system 33.61% –
Access to hospital 7.00% –
Access to health facilities 37.50% –
Household expenditure in mass media 37.811 9.017
Number of motor vehicles 0.29 0.671
Number of bicycles 0.31 0.699
Number of radios 1.03 0.858
Number of fridges 0.39 0.616
Number of stoves 0.30 0.480
Source: 1993 South Africa Integrated Household Survey.
621South African Journal of Economics Vol. 76:4 December 2008
© 2008 The Authors.

Journal compilation © Economic Society of South Africa 2008.
months and 6 years of age. The first three specifications of Table 4 show OLS estimates
of equation (5) where the first specification is a benchmark OLS regression of child’s
health (measured by height for age Z-score) on years of mother’s education controlling for
basic child and mother characteristics such as race, gender and weight of the child, age of
child and mother; and if the mother is a widow and if father of the child is absent. As seen
from the table, one more year of mother’s education has a positive and statistically
significant (1%) effect of approximately 0.049 standard deviations on average on a child’s
height for age.
However, the estimator of the mother’s education coefficient in the benchmark can
suffer from omitted variables bias as discussed above. This justifies controlling for
household characteristics such as fathers education, disposable household income,
number of siblings and the quality of the house they live in captured by number of rooms,
household usage of electricity and tap water; and availability of a toilet. The second
specification in Table 4 reports this regression. Moreover, community characteristics
are also called for, such as whether the household resides in an urban area and if the
community in which the household resides has health facilities. As seen, the coefficient on
years of mother’s education is reduced in half but still has a positive and statistically
significant (1%) effect of approximately 0.026 standard deviations on average on a child’s
Table 4. Dependent variable is children’s height for age Z-score
OLS estimates 2SLS estimates
I II III IV V VI
Mother’s education 0.049 0.026 0.020 0.147 0.151 0.137
(0.008)** (0.009)** (0.011)
†
(0.038)** (0.087)+ (0.091)
†
White (= 1 if child is White) 0.763 0.318 0.310 0.359 0.415 0.358
(0.143)** (0.144)* (0.144)* (0.188)+ (0.262) (0.277)
Female (= 1 if child is a girl) 0.229 0.220 0.219 0.228 0.225 0.252

(0.064)** (0.065)** (0.065)** (0.065)** (0.066)** (0.071)**
Age in months -0.061 -0.059 -0.059 -0.058 -0.058 -0.062
(0.012)** (0.012)** (0.012)** (0.012)** (0.012)** (0.013)**
Age in months squared 0.000 0.000 0.000 0.000 0.000 0.000
(0.000)** (0.000)** (0.000)** (0.000)** (0.000)** (0.000)**
Weight of child 0.188 0.183 0.183 0.180 0.181 0.185
(0.055)** (0.055)** (0.055)** (0.054)** (0.055)** (0.056)**
Mother’s age 0.015 0.013 0.013 0.028 0.026 0.018
(0.005)** (0.005)** (0.005)** (0.006)** (0.009)** (0.012)
Widow (= 1 if mother of child is a widow) -0.001 0.039 0.050 0.004 0.013 -0.155
(0.136) (0.139) (0.139) (0.138) (0.164) (0.237)
Absent (= 1 if father of child is absent) -0.147 -0.088 -0.085 -0.210 -0.217 -0.191
(0.066)* (0.067) (0.067) (0.075)** (0.127)+ (0.137)
Father’s education 0.007 0.006 -0.027 0.072
(0.011) (0.011) (0.081) (0.120)
Household disposable income 0.001 0.001 -0.001 0.000
(0.000)* (0.000)* (0.003) (0.003)
H = 1 if hh resides in community with a hospital -0.194 -0.042 -0.177 -0.348
(0.126) (0.294) (0.130) -2,871
OHF = 1 if hh resides in community with other
health facilities
0.049 -0.105 0.039 2,110
(0.064) (0.136) (0.070) -2,064
Mother’s educ ¥ H -0.019 0.036
(0.031) (0.335)
Mother’s educ ¥ OHF 0.020 -0.277
(0.016) (0.275)
Comunity and household characteristics No Yes Yes No Yes Yes
Overidentification test of IV list (p-value) – – – 0.60 0.63 0.46
N 3,919 3,912 3,912 3,919 3,912 3,912

R
2
0.14 0.15 0.15 – – –
Robust standard errors in parentheses.
+ Significant at 10%; * significant at 5%; ** significant at 1%;
†
significant at 10% one tail.
Source: 1993 South Africa Integrated Household Survey.
South African Journal of Economics Vol. 76:4 December 2008622
© 2008 The Authors.
Journal compilation © Economic Society of South Africa 2008.
height for age for every year of formal education. Intuitively, increasing mothers’
education by one standard deviation (4 years) would imply that on average a 2-year-old
child would be approximately 0.5 centimeters taller.
The third specification in Table 4 includes interaction terms between MED and health
facilities to assess the impact of the efficiency channel in the usage of health inputs.
The coefficients of the interaction terms are not statistically significant. The lack of
significance could be due to the cost of using these health inputs. Even if health facilities
are available, mothers may choose not to use them for various reasons such as
transportation costs and high monetary costs. Moreover, the estimators of the interaction
coefficients can be biased due to the potential endogeneity of MED.
In all three OLS specifications of Table 4 one can observe that heavier children tend to
have a higher height for age Z-score. This is not surprising since both measures are clearly
positively correlated. It is interesting to notice that girls tend to have better health status
relative to boys which is statistically significant while White South African children are
relatively healthier than children from other races in the country. Furthermore, we find
that the child age coefficient is negative and significant. This result by itself implies that
as children get older, differences in the height for age measures for American and South
African children increase. Since the coefficient on child’s age squared is positive and
significant it appears that the negative impact of age on child’s health increases in a

decreasing rate.
When analysing the impact that households’ characteristics have on children’s health
the first thing to notice is that even though household disposable income is positive, its
magnitude is quite small once mother’s and father’s education are controlled for. This
result goes in line with previous findings such as Thomas et al. (1991). Interestingly,
children with more educated fathers and who lived in wealthier households do not seem
to have a better health status. Moreover, children that had a younger mother and an
absent father had a relatively poorer health status in a statistically significant way. It’s not
the same to be a mother at age 16 than at age 30, where the women are more mature and
physically better prepared. Finally, in the second specification of Table 4 unhealthier
children seem to reside in communities with hospitals and other health facilities which
clearly cannot be interpreted as a causal effect since it is possible that these facilities could
suffer of endogenous placement policies.
2SLS Results
As discussed above mother’s education, father’s education and disposable household
income could be endogenous. Under this scenario the last three columns of Table 4 report
two-stage least square (2SLS) estimates of the respectively three OLS specifications. In the
fourth specification of Table 4 mother’s education is the only endogenous variable and
is instrumented by the following variables: exposure of the household to mass media,
expenditure in newspapers and entertainment, if the main language in the household is
Afrikaans or English and whether someone in the household has moved or migrated for
academic reasons. The estimate coefficient on MED is positive and statistically significant
but with a much greater magnitude compared to the first OLS specification.
12
Interpreted
12
Even though the first stage regression is not reported the four instrumental variables are jointly
significant at the 1%.
623South African Journal of Economics Vol. 76:4 December 2008
© 2008 The Authors.

Journal compilation © Economic Society of South Africa 2008.
causally one more year of a mother’s education increases approximately 0.14 standard
deviations on average a child’s height for age.
When household and community characteristics are included as in the fifth
specification of Table 4, where MED as well as FED and household disposable income are
instrumented, the magnitude of the MED coefficient remains similar but notice that
standard errors increase due possibly to both the 2SLS procedure and multicollinearity
of the three endogenous variables. The first stage of this 2SLS procedure is reported
in Table 5 where a robust Lagrange Multiplier statistic to test joint significance of the
relevant instruments of each endogenous variable is reported. In all cases the relevant
instruments are jointly significant at 1%.
13
The last column of Table 4 reports 2SLS
estimates when the interaction terms of MED with health facilities available at the
community level are included. The coefficients of the interaction terms are again not
statistically significant.
Moreover, it is worth mentioning that all 2SLS estimates satisfied an over identification
test for the list of instrumental variables. For any specification the null of joint exogeneity
of the instrumental variables is not rejected at even 40% of statistical significance. This
gives confidence on the exclusion assumptions necessary for the consistency of the 2SLS
estimator of b
1
.
Under these results, our preferred specification is the fifth one in Table 4. As can be
observed in this model the 2SLS estimates maintain the same sign as the OLS estimates.
13
In this specification the instruments for mother’s education are media and expenditure in
newspapers and/or media; while for father’s education the corresponding instruments are studying
away and English/Afrikaans as household language; finally, the instruments for disposable income
correspond to the rest of the variables (i.e. number of motor vehicles, stoves, television sets,

telephones and fridges owned by the household).
Table 5. OLS first stage regressions
Mother’s educ. Dep variable
father’s educ.
Disp. hh
income
English or Afrikaan is language of household -0.522 0.739 -11,765
(0.202)** (0.172)** (4.012)**
Media (= 1 if household buys newspaper and/or entertainment media) 0.589 0.337 21,048
(0.129)** (0.115)** (2.566)**
Expenditure in newspapers and/or media 0.001 0.002 0.127
(0.001) (0.002) (0.064)*
Studying away(= 1 if household has member studying away) 4,162 0.609 410,930
(0.311)** (0.275)* (5.654)**
Own house (= 1 if hh resides in own house) 0.022 -0.084 -6,339
(0.141) (0.126) (2.439)**
Number of motor vehicles 0.402 0.653 7,273
(0.115)** (0.112)** (2.850)*
Number of TV sets 0.560 0.267 9,544
(0.121)** (0.113)* (2.930)**
Number of telephones -0.179 0.069 2,274
(0.254) (0.161) -4,197
Number of stoves 0.267 0.297 7,571
(0.192) (0.163)+ (3.838)*
Number of fridges 0.537 0.083 1992
(0.131)** (0.124) -3,076
Includes all other variables in second stage Yes Yes Yes
Robust LM test for joint significance of relevant instruments (p-value) 0.0001 0.0001 0.0001
N 3,912 3,912 3,912
R

2
0.29 0.34 0.49
Robust standard errors in parentheses.
+ Significant at 10%; * significant at 5%; ** significant at 1%.
South African Journal of Economics Vol. 76:4 December 2008624
© 2008 The Authors.
Journal compilation © Economic Society of South Africa 2008.
As is well known a 2SLS estimation procedure increases standard errors; nonetheless,
mother’s education is statistically significant at 10%. Importantly, if we believe that the
parameter of interest b is estimated consistently then the magnitude of this estimate
increases strongly: one more year of a mother’s education increases on average the height
for age Z-score of a child in 0.15 standard deviations. This finding is similar to that of
Glewwe (1999) in which after instrumenting maternal education and disposable income
the magnitude of the coefficient of interest significantly increased. Compared to the OLS
case, the 2SLS coefficient suggests that increasing mothers’ education by one standard
deviation (4 years) would imply that on average a 2-year-old child would be
approximately 3.5 centimeters taller. Although this finding seems quite large, the result is
quite similar to that reported in Chen and Li (2006) for Chinese adopted children: on
average a Chinese woman with 4 more years of education would increase the height of the
adopted child in 3 centimeters.
Finally, we explored with the second and fifth specifications of Table 4 the differential
impact of mother’s education on children’s health depending on their age. The medical
and public health literature shows that young children (ages 0 to 2) undergo a critical
transition period, namely they have underdeveloped immune systems and are relatively
more vulnerable to infections and disease. This age period is characterised by fast growth
that coincides with a changing diet from breast milk to prepared foods. At a certain time,
breastfeeding alone becomes inadequate for their nutrient requirements which can
generate trauma and stress in their transition to prepared foods if the diet deteriorates in
quality and perhaps in quantity. It is not surprising therefore that child’s health outcomes
have generally exhibited a decline from 6 months of age through the second year of life,

followed by a turnover and a continuous improvement thereafter.
14
This suggests a U
shape pattern in the transition period from breastfeeding intakes to prepared food intakes.
Following the U shape pattern idea, the sample was divided in two age groups of
children: those between 6 months and 24 months of age and those between 24 and 72
months of age. We conjecture that children between 6 and 24 months, compared to older
ones, are more dependent on their mother’s presence (e.g. breast feeding) but not on her
education level. Hence, the coefficient on MED should be smaller than those obtained
using the full sample of children. The results found are summarised in Table 6 for both
OLS and 2SLS estimations. We find that MED is important for older children while not
for younger ones which is in line with the literature. Interestingly, the OLS and 2SLS for
the older aged group is similar in practical terms to the findings for the whole sample
which supports the conjecture that mother’s education is important only for children that
do not breastfeed and which need well prepared food.
6. CONCLUSIONS
Several studies in different countries have found that maternal education affects positively
and significantly child’s health outcomes. In this paper we show that South Africa is no
exception to that rule. Children with more educated mothers have a better long-term
relative health status measured by height for age Z-score.The 2SLS coefficient suggests that
increasing mothers’ education by one standard deviation (4 years) would imply that on
average a 2-year-old child would be approximately 3.5 centimeters taller. This finding is
14
Barrera [1990].
625South African Journal of Economics Vol. 76:4 December 2008
© 2008 The Authors.
Journal compilation © Economic Society of South Africa 2008.
important since, as mentioned, previous studies have found that healthier status during
childhood are related to better education and labour market outcomes in adulthood.
Across the different specifications considered we gradually controlled for several

channels through which mother’s education can affect child’s health. The objective was
to isolate the knowledge channel in order to assess the reallocation of resources due to
better knowledge and information. Several interesting results emerge from here. Father’s
education is not important once mother’s education is controlled for. Secondly, neither
health facilities nor their interaction with mother’s education appear to have a positive
significant effect on the height for age of the children between 6 months and 6 years of
age in South Africa. The intuition is that even if health facilities are available to the
households these may choose not to use them if the cost of doing so is too high. The two
results have important policy implications since they suggest that governments should
invest the scarce resources primarily in women’s education and perhaps subsidise the use
of local health facilities for the poorer households.
Finally, following the medical literature that reports a U pattern in children’s health we
divided the sample in two different age groups: those between 6 and 24 months of age and
those between 24 and 72 months. We find that while mother’s education is not important
during the breastfeeding period of the child, it becomes very relevant once the child is
older than 2 years of age.
REFERENCES
BARRERA, A. (1990). The role of maternal schooling and its interaction with public health programs in child health
production. Journal of Development Economics, 32: 69-91.
BEHRMAN, J. and DEOLALIKAR, A. (1988). Health and nutrition. Handbook of Development Economics,V1.
Table 6. Dependent variable is height for age Z-score
Յ 24 months >24 months
I II III IV
OLS 2SLS OLS 2SLS
Mother’s education 0.010 0.073 0.027 0.157
(0.017) (0.177) (0.010)** (0.093)+
Father’s education 0.013 -0.135 0.006 0.010
(0.024) (0.147) (0.012) (0.083)
Household disposable income 0.001 0.004 0.001 -0.003
(0.001) (0.006) (0.001)* (0.002)

White (= 1 if child is White) 0.315 0.454 0.316 0.412
(0.267) (0.646) (0.166)+ (0.263)
Female (= 1 if child is a girl) 0.441 0.451 0.142 0.136
(0.107)** (0.123)** (0.073)+ (0.077)+
Age in months -0.229 -0.205 -0.044 -0.044
(0.065)** (0.072)** (0.022)* (0.022)*
Age in months squared 0.003 0.002 0.000 0.000
(0.002) (0.002) (0.000) (0.000)
Weight of child 0.512 0.505 0.135 0.135
(0.047)** (0.049)** (0.051)** (0.051)**
Mother’s age 0.011 0.010 0.016 0.030
(0.009) (0.021) (0.005)** (0.009)**
Widow (= 1 if mother of child is a widow) 0.544 0.653 -0.071 -0.138
(0.350) (0.411) (0.147) (0.178)
Absent (= 1 if father of child is absent) -0.062 -0.014 -0.072 -0.268
(0.128) (0.261) (0.075) (0.128)*
Observations 1,181 1,181 2,731 2,731
R
2
0.28 – 0.14 –
Robust standard errors in parentheses.
+ Significant at 10%; * significant at 5%; ** significant at 1%.
South African Journal of Economics Vol. 76:4 December 2008626
© 2008 The Authors.
Journal compilation © Economic Society of South Africa 2008.
BEHRMAN, J. and ROSENZWEIG, M. (2004). Returns to birthweight. Review of Economics and Statistics, 86(2):
586-601.
CASE, A., FERTIG, A. and PAXSON, C. (2003). From cradle to grave? The lasting impact of childhood health and
circumstance. NBER Working Paper 9788.
CHEN, Y. and LI, H. (2006). Mother’s education and child health: Is there a nurturing effect? Working Paper.

CURRIE, J. and HYSON, R. (1999). Is the impact of health shocks cushioned by socioeconomic status? The case of low
birthweight. American Economic Review Papers and Proceedings, 89(2): 245-250.
DUFLO, E. (2000). Grandmothers and granddaughters: Old age pension and intra-household allocation in South Africa.
World Bank Economic Review, 17(1): 1-25.
GLEWWE, P. (1999). Why does mother’s schooling raise child health in developing countries? Evidence from Morocco.
Journal of Human Resources, 34(1): 124-159.
KESWELL, M. (2004). Education and racial inequality in post Apartheid South Africa. Santa Fe Institute Working Paper
04-02-008.
LAM, D. (1999). Generating extreme inequality: Schooling, earnings, and integrated transmission of human capital in
South Africa and Brazil. Population Studies Center, Research Report, No. 99-439.
STRAUSS, J. (1990). Households, communities, and preschool children’s nutrition outcomes: Evidence from rural Côte
d’Ivore. Economic Development and Cultural Change, 38(2): 197-234.
STRAUSS, J. and THOMAS, F. (1995). Human resources: Empirical modeling of household and family decisions.
Handbook of Development Economics,V3.
THOMAS, D. (1994). Like father like son or like mother like daughter: Parental education and child health. Journal of
Human Resources, 29(4): 950-989.
———, STRAUSS, J. and HENRIQUES, M. H.(1991). How does mother’s education affect child height? Journal of
Human Resources, 26(2): 183-211.
WOLFE, B. and BEHRMAN, J. (1987). Women’s schooling and children’s health: Are the effects robust with adult sibling
control for the women’s childhood background? Journal of Health Economics, 6(3): 239-254.
627South African Journal of Economics Vol. 76:4 December 2008
© 2008 The Authors.
Journal compilation © Economic Society of South Africa 2008.