1
Malnutrition and Inequality in Ecuador
Ana Larrea
Abstract
We measure the effect of inequality on stunting in children (<5) by regressing the z-score of height-for-
age using multilevel regressions; and the dummy for stunting using logistical multilevel regressions
against the provincial, county and parish Gini coefficients while controlling for individual, maternal,
household and contextual characteristics. We find the Gini coefficient has a significant deleterious effect
on both z-score and stunting except in the logistic parish model although data on nutritional status of
parents is lacking. In conclusion, inequality is detrimental to children’s nutrition; it may increases pre-
natal maternal stress or maternal agency over household resources affecting child’s health.
2
1. Introduction
Chronic malnutrition affects 1 in 4 children in the world (De Onis, et al., 2012), and is the root cause of
just under half (45%) of deaths in children under the age of five (Horton & Lo, 2013). Its effects are
potentially long-term and create deficits in cognition and educational achievements (Granthan-
MacGregor, et al., 2007; Grantham-McGregor, et al., 2000; Walker, et al., 2000; Walker, et al., 2007)
playing a key role in the intergenerational transmission of poverty. It is an important public health
problem in many Latin American countries, particularly among indigenous populations in countries with
strong socioeconomic disparities such as Bolivia, Peru, Ecuador, Guatemala and Honduras (Larrea &
Freire, 2002; Farrow, et al., 2005).
There are two immediate causes of this type of malnutrition, fisrtly, insufficient access to nutritents and
secondly, high disease exposure (Larrea & Freire, 2002). Biological mechanisms determine malnutrition
at an individual level, however, the “lifestyles” and “behaviours” that lead to both a reduced nutritional
intake and a high level of disease exposure may, in fact, be shaped and contrained by socioeconomic
context and regional disparities at the aggregate level (Larrea & Freire, 2002; Diez-Roux, 1998).
In this study, we measure the effect of inequality on chronic child malnutrition (stunting) in Ecuador. We
use OLS and multilevel models to regress the Gini coefficient measured at the local, county and
provincial level against the z-score of height-for-age. Additionally, we use logistical multilevel models to
regress the same Gini coefficients against a malnourishment dummy while, in both cases, controlling for
household consumption per capita. In both models we control for the characteristics of the child, the
mother, the household, and other contextual variables including but not limited to access to healthcare,
education, employment conditions, prenatal and postnatal care, adequate housing, diet and add fixed
effects for ethnicity. By including the consumption variable measured at the household level and not at
the community level we are able to control for the effect inequality independently of individual income.
Our results show that, in every OLS, multilevel and logistical multilevel model, the Gini coefficient has a
significant deleterious effect on both the z-score of height-for-age and the probability of being
malnourished, except for the logistical parish (local) level model. Our income control variable, household
consumption per capita, is also significant and has a beneficial effect on both dependent variables in every
model. Therefore, both variables are relevant independent of each other.
As far as social determinants of health go, inequality has generated some debate regarding its relevance
independent of income. The debate constructs inequality and income as two sides of the same coin,
arguing that inequality is relevant because income is relevant (Deaton, 2003) or when income cannot
explain the variance in health (Preston, 1975; Lynch, et al., 2004). We argue that there is a lot to say on
3
the inherent effects of inequality, independently of income (Wilkinson, 2000). In other words, whichever
effect income may have on child nutrition, inequality has an additional independent effect inherit to the
characteristics of “unequal” vs. “equal” societies.
In order to explain our findings we propose that inequality will affect individual health psychosocial
stress due to social status and violence due a lack of social cohesion (Wilkinson, 1996; Ellison, 2002;
Macinki et al., 2003). We maintain that the experience of living on a low income in societies with high
levels of inequality is different than it is in societies with low levels of inequality. We argue that a highly
unequal context may lead to increased anxiety and chronic stress due to feelings of exclusion, shame and
mistrust which may arise among those least advantaged (Lynch, et al., 2004; Wilkinson, 2000; Davey
Smith & Egger, 1996; Lynch, et al., 2000; Lynch, et al., 2000; Lynch, et al., 2001) as well as increased
violence against women (VanderEnde, et al., 2012) although we do not test this directly.
Our first pathway is based on the evidence that chronic amounts of stress among pregnant women
increases the levels of CRH,1 which regulates fetal maturation and increases the risk of low birth weight
(LBW) (Beydoun & Saftlas, 2008; Camacho, 2008; Mansour and Rees, 2011). LBW is an important
determinant of chronic child malnutrition (Marins & Almeida, 2002; Willey et. al., 2009; Aerts, et al.,
2004; El Taguri, et al., 2009; Adair & David, 1997). Our second pathway is related to the eroding effect
that inequality may have on social cohesion (Wilkinson, 2000; Kawachi, et al., 1999; Wilkinson, 1996)
which may be associated with an erosion of the interpersonal empowerment of women at the community
level (Peterson & Hughey, 2004; Speer, et al., 2001) affecting the mothers agency over household
resources (Power, 2006; Satatistics Canada, Candian Centre for Justice Statistics, 2000; Lupri, 1990;
Antai, et al., 2014) which may refocus these resources away from the health and nutritional needs of
children (Power, 2006; Charles & Kerr, 1988; DeVault, 1991; Maitra, 2004; Shroff, et al., 2009; Imai, et
al., 2014; Ackerson & Subramanian, 2008).
In a recent review of 98 studies on the effect of inequality on health (Lynch, et al., 2004), there are
unexplored issues which we believe this article addresses. Fisrtly, there is only a small percentaje of
studies which focus on child health (23.7%). Most of these studies measure Infant Mortality, which unlike
nutrition, does not play a role in the intergenerational transmition of poverty. Secondly, only 10.2% of the
studies include Latin American countries which is shortfall given it is one of the most unequal regions of
the world (Inter-American Development Bank, 2000) and perhaps the ideal testing ground for the effect
of inequality. Finally, only 9.2% of the studies measure the Gini coefficients at different levels. In this
study, we find that the magnitud of the Gini coefficient is smaller as the areas over which it is measured
decreases in size. Therefore, in order to fully assess the impact of inequality it is important to take
1 Croticotrophin-Releasing Hormone
4
different levels of aggregation as we have done. This study expands on the evidence first put forward by
Larrea and Kawachi (2005).
Being exposed to malnutrition during infancy is closely related to poor cognitive and educational
performance. There is considerable evidence that reduced protein-energy malnutrition is associated with
deficits in cognition and school achievements (Grantham-McGregor, et al., 2000). Granthan-MacGregor,
et al. (2007) find2 that children who where currently stunted (chronically malnourished) were less likely
to be enrolled in school, more likely to enroll late, attain lower achievement levels or grades and have
poorer cognitive abilities. In longitudinal studies they found that stunting predicted age of walking, later
cognition and/or school progress and was related to literacy, numeracy, grade repetition, dropouts and
intelligent quotient (IQ) later in life. They argue that children who do not reach their developmental
potential are less likely to become productive adults because they have fewer years of education and
because they are learning less per year of schooling. They estimate that the loss in adult yearly income
from being stunted (chronically malnourished) is 22.2%, assuming an increase in yearly income of 9%
per year of schooling (Granthan-MacGregor, et al., 2007). Other authors have also found that growth-
restricted children have significantly poorer performance than non-growth-restricted children on a large
range of cognitive tests. Their results support the conclusion that growth restriction has long-term
functional consequences (Walker, et al., 2000).
2. Literary review
In this section we will attempt to give the reader a well-rounded understanding of the health-inequality
relation. Firstly, in order to provide a conceptual framework, we review the theorized functional relation
between income, inequality and health. Secondly, so as to provide the context into which this study will
fall and how it contributes, we review the empirical evidence of the effect of inequality. Thirdly, in order
to present an understanding of the inherit effects of inequality and how public policy might be formulated,
we provide the pathways through which we believe the nutrition-inequality relation runs.
i. The functional relation
Wafstaff and Van Doorslaer (2000) research the relation between health and income distribution, in this
paper we will focus on two of these hypotheses: the Absolute Income Hypothesis (AIH) and the Income
Inequality Hypothesis (IIH). The AIH is the notion that individual health is a concave function of
individual income. Therefore, at the individual level, each additional dollar of income raises individual
health by progressively smaller amounts. At the country level, average health will improve as average
income increases. The decreasing tendency of the AIH implies that there is a threshold beyond which the
2 In their study on developmental potential in the developing word.
5
association between income and health weakens. The relationship was tested by Lynch at al. (2004) for
life expectancy. Among richer countries, the strength of said association depends crucially on the country
and the time period (Lynch, 2000; Lynch, et al., 2000; Lynch, et al., 2001). The resulting unexplained
variation in average health within these richer countries led to the reasoning that if it is not average
income, then perhaps it is the distribution of income within a country which helps explain the variation in
life expectancy (Preston, 1975; Lynch, et al., 2004). This argument led to a greater focus of the empirical
research on the effect of inequality on health.
The notion that individual health is a decreasing function of income inequality is known as the IIH.
Theoretically, a curvilinear association between income and health at the individual-level is sufficient to
produce and aggregate-level association between income inequality and the average health of a
population (Rodgers, 1979).3 This implies that redistribution within that population could improve its
average health through the health purchasing power of individual-level income rather than through any
process inherit to inequality per se. This is an argument generally used to discredit ecological studies
(dependent and independent variables are aggregated) which explore the inequality-health relation stating
that average health is only affected by inequality through the income effect of redistribution. However,
this does not explain the significance of inequality on health outcomes in multilevel studies (dependent
variable is individual and independent variable is aggregated) and it does not address the psychosocial
effects inequality has on individual’s health (Wilkinson, 2000).
In this study we have built a multilevel model in which the dependent variable is measured at the
individual level not at an aggregated level, therefore any effect that income may have on individual health
is controlled by the per-capita household consumption variable. This allows us to isolate the effect
inequality has on individual health independent of any income driven effect that distribution may have on
average health.
ii. The empirical evidence
Wafstaff and Van Doorslaer (2000) find ample and strong empirical support for the AIH and some
evidence for the IIH. They observe that the strength of the effect of inequality depends crucially on how
well other influences on health are controlled for, especially individual income and those which vary
systematically over geographical areas and time.
3 Why does this happen? The idea is that, given the income-health relation is convex, if we take x income from one
person and give it to another, the former becomes poorer and the latter becomes richer. In this case, inequality will
increase, however average income will stay the same. As a consequence, the health of the former will deteriorate
more, in magnitude, than the improvement in the health of the latter. Therefore, average health will decrease even as
average income is constant.
6
Macinko et al. (2003) present a review of 45 studies where they find 33 (73.3%) have a significant
inequality health relation, and 12 (26.6%) have a non-significant relation. Lynch et al. (2004) presented a
review of 98 studies on inequality and health of which 42% found that all measures of association showed
statistically significant relationships between smaller income difference and better health, another 25%
were partially supportive and 33% provided no support. Wilkinson & Pickett (2006) reviewed 168 studies
and found that 52% had supportive evidence, 25% had partially supportive evidence and 22% found no
evidence of the association between income inequality and health. Larrea and Kawachi (2005), the study
on which this paper is based, propose that the lack of association between income inequality and health in
wealthy countries is due to a threshold effect. This idea, developed by Subramanian and Kawachi (2004),
is that there might be a threshold beyond which inequality is too low to have a significant effect on health.
This idea is based on the fact that studies conducted outside the United States have generally failed to find
an inequality-health association. However, almost all the non-US countries listed in these studies are
considerably more egalitarian in their distribution of income than the United States. Also, when there are
cases of relatively more unequal countries there is some support for the relation (Subramanian &
Kawachi, 2004).
This study contributes to the debate on various fronts. In the Lynch et al. (2004) review we found 10
(10.2%) out of the 93 studies4 are based on or include Latin American countries. We believe research in
this regions is relevant given it is the most unequal region in the world and therefore is an ideal testing
ground for the hypothesis (Larrea & Kawachi, 2005, Inter-American Development Bank, 2000).
Likewise, 23 (23.71%) out of the 97 studies5 are based on children’s health. We believe that research on
children’s health is somewhat lacking given most of the studies performed focus on infant mortality (IM).
IM does not play a role in the intergeneration transition of poverty in the way that stunting of chronic
malnutrition does. Finally, 9 (9.27%) out of the 97 studies with information, measure the Gini coefficient
for more than one geographic area of aggregation6 in order to lest the inequality health relation. Most of
the studies included in the review measure inequality on only one level of aggregation. This does not
allow us to see how the effect of inequality changes if it is measured over large vs small areas. Given the
effects of the Gini coefficient may vary depending on the how we define the area of interest, measuring
the Gini coefficient at various levels of aggregations allows a more profound analysis of inequality on
health. In the Wilkinson and Pickett review (2006) there are 45 international studies, 58 state levels
studies, 25 county level studies and 40 small area level studies. As we can, see there are relatively less
studies performed at the county level while mostly studies focus on the state level. Additionally, the
4 In which the information was provided. 5 In which the information was provided. 6 For example, not only state or municipality or census tract, but two or more of them in one study.
7
distribution of the supportive, mixed and unsupportive evidence favours the supportive at the state level
(S: 51.7%, M: 25.8%, U: 22.4%) while it is fairly balanced at the small area level (S: 30%, M: 35%, U:
35%). This demonstrates the importance of including various levels of aggregation given the effect that
this has on the results.
Larrea and Kawachi (2005) on which this paper is based, contributes in these three fronts, however, we
feel that his paper improves on theirs given our study uses not only multilevel but also logistical
multilevel regressions as well as finds a significant relation between the Gini coefficient and malnutrition
at the parish and county level which was not the case for Larrea and Kawachi (2005). This may be due to
the smaller sample size (2723 children) of the Living Standards Measurement Survey (LSMS) they used
(1998) given our LSMS (2006) sample is larger (6003 children). Finally, we also provide a theoretical
framework which explains the pathways through which inequality may be affecting individual child
nutrition based on the psychosocial effects of inequality (Wilkinson, 2000).
iii. The pathways
It is proposed by Wilkinson (1996) and Kennedy & Kawachi (1999) and outlined in Ellison (2002) and
Macinko et al. (2003) that inequality may affect individual health on three levels:
(i) Individual level: inequality increases perceptions of injustice and exclusion for individuals
increasing psychological and psychosomatic stress – social status.
(ii) Community level: inequality fosters distrust and inter-personal violence (perhaps intimate
partner violence) psychosocial stress – social cohesion.
(iii) Structural level: inequality reflects sociopolitical system with inadequate redistributive
policies and safety net provisions.
The structural level does no actually connect inequality to health directly. Rather, at this level we see the
unequal distribution of income as a parallel phenomenon to the unequal distribution of social services
relevant to individual health and nutrition (Ellison, 2002). Lynch et al. (2000) argues that income
inequality is symptomatic of a lack of resources at the public level. That is to say, inequality reflects an
under-investment in health and social infrastructure given it is the result of a historical, political and
economic process which has influenced the nature and availability of health supportive infrastructure.
This process shapes the structural matrix of contemporary life which likely influences individual health,
particularly of those who have fewer resources. However, this does not imply that inequality is the cause
of an inadequate redistribution system but rather another one of its outcomes. An underinvestment in
social goods, which may lead to an unequal distribution of healthcare and education, has an effect on
individual health; however inequality of income does not imply inequality of social goods.
8
Notwithstanding, the at the structural level, income inequality may actually be a manifestation of the
unequal structural conditions that do affect health (Davey Smith & Egger, 1996; Lynch, et al., 2000;
Lynch, et al., 2001; Lynch, et al., 2004). Given that, at the structural level, the health effects of material
conditions are historically contingent and disease specific, we do not consider this a pathway per se
(Lynch, et al., 2000).
In this paper we propose two pathways: prenatal maternal stress and maternal agency over household
resources. As we can see, both pathways between inequality and children’s health are realized through the
health and wellbeing of the mother. The mechanisms through which the pathways operate are complex so
we have conceptualized them in stages.
In the pre-natal maternal stress pathway we have three stages: (1) the psychosocial mechanism, by which
relatively high inequality increases chronic stress; (2) psycho neuroendocrine mechanism, by which
chronic stress affects fetal maturation and is conducive to low weight at birth; (3) the statistically
significant relation between low birthweight and chronic malnutrition during infancy.
The maternal household violence mechanism has four stages: (1) the relation between social cohesion and
interpersonal empowerment, by which women’s interpersonal empowerment weakens as social cohesion
is eroded; (2) a reduction in women’s empowerment, may reduce a woman’s agency over household
resources; (3) a reduction in maternal agency over household resources may lead to a focus of these
resources away from the needs of the children, enabling an increase in chronic malnutrition.
These pathways show how inequality may affect health through the individual level mechanism by
affecting the social status of women, and through the community level by eroding social cohesion
increasing stress and reducing women’s interpersonal empowerment. We believe these pathways are
relevant given there is evidence that intrauterine growth restriction, maternal depression, and exposure to
violence among other things are important risk factors on child development (Walker, et al., 2007). In this
section we will explain in detail the two pathways we propose above.
a. The prenatal maternal stress pathway
The recognition of the importance of psychosocial factors, specifically channeled through chronic stress is
a crucial development in our understanding of the social determinants of health and has received greater
attention as a source of chronic stress (Berkman and Glass, 2000; Marmot, 2004; Marmot and Wilkinson,
2005). A social context in where there are strict social hierarchies which generate a strong perception of
place and station, common in relatively unequal societies, may alienate some people and produce negative
emotions among them, such as shame or mistrust (Lynch, et al., 2004). When these emotions are
translated into antisocial behavior, a reduction in social capital and an increase in stress, we refer to this
9
process as the psychosocial mechanism resulting in a decline in health. The way in which stress affects
individual health is known as the psycho-neuroendocrine mechanism. Individual health may also be
affected by stress-induced behaviors such as smoking (Wilkinson, 2000). This hypothesis has been
researched in major review papers (Nguyen & Peschard, 2003) and has been the subject of various
international and governmental organizations reports (Agren, 2003; Bennett, 2003; Health Canada, 1999;
Howden-Chapman & Tobias, 2000; Organization of Economic Cooperation and Development, 2001;
Persson, et al., 2001; Turrell, et al., 1999) which demonstrates its broader acceptance within the research
and policy community (Lynch, et al., 2004).
When we have chronic amounts of stress, immunity is down-regulated, there is potential wear on the
cardiovascular system, and we are vulnerable to a wide range of health problems (Wilkinson & Pickett,
2009). The medical literature indicates that pre-natal stress increases levels of CRH7, which regulates the
duration of pregnancy and fetal maturation and thus increases the risk of adverse birth outcomes
(Beydoun & Saftlas, 2008, Mansour and Rees, 2011, Camacho, 2008). Beydoun and Saftlas (2008), in
their review of the literature on the effect of pre-natal maternal stress (PNMS) on fetal growth, find that
nine out of 10 studies report significant effects of PNMS on birth weight, low birth weight (LBW) or fetal
growth restriction. However, they also find that the evidence was predominantly derived from animal
studies. Nevertheless, and despite methodological study limitations, the overall evidence is revealing of
an independent association between PNMS and numerous physical and mental health outcomes (Beydoun
& Saftlas, 2008). Almond and Currie (2011) find numerous studies providing evidence of the long-term
consequences of a wide variety of intrauterine shocks (Almond & Currie, 2011). For example,
intrauterine exposure to a terrorist attack or severe famine, even psychological stress experienced in the
first trimester of pregnancy, can reduce birth weight (Camacho, 2008; Eskenazi et al., 2007). Camacho
(2008) finds that the intensity of random landmine explosions during a woman’s first trimester of
pregnancy has a significant negative impact on child birth weight. This finding persists when mother
fixed effects are included, suggesting that neither observable nor unobservable characteristics of the
mother are driving the results (Camacho, 2008). Insofar as we have argued inequality is associated with
stress and pre-natal maternal stress with LBW, Kaplan et al. (1996) demonstrate the link is measurable
and produce evidence that income inequality in the United States was significantly associated with rates
of low birth weight.
There are various studies which find that children with LBW are at higher risk to suffer chronic
malnutrition. Marins and Almeida (2002) find in Niterói, Brazil that LBW and low family income could
be characterized as important under-nutrition risk factors both for the “0-12 months” and for the “above
7 Croticotrophin-Releasing Hormone
10
13 months” age ranges. They argue that, a child’s birth weight deficits appear to have effects on a child’s
growth that extend for years after birth (Marins & Almeida, 2002). Willey et. al. (2009) found, in
Johannesburg and Soweto, that increased likelihood of stunting (chronic malnutrition) was seen in LBW
children (Willey et. al., 2009). Aerts, et al. (2004) perform a cross-sectional population-based study of
determinants of growth retardation (chronic malnutrition) in under five children in Porto Alegre, Brazil.
The main determinants were LBW, per-capita family income, among others (Aerts, et al., 2004). Taguri et
al. (2009), who used a multivariate analysis in order to ascertain predictors of stunting in children under
five in Libya, found that risk factors were mainly LBW, as well as other variables like paternal education
and age (El Taguri, et al., 2009). Adair and David (2007) used a multivariate discrete time hazard model
to estimate the likelihood of becoming stunted in each two month intervals and found that the likelihood
of stunting was significantly increased by LBW and other variables such as febrile respiratory infections
and early supplemental feeding. Additionally, they find that the effect of birth weight was strongest in the
first year and that breast-feeding, preventive health care and taller maternal stature significantly decreased
the likelihood of stunting. (Adair & David, 1997). In addition, birth weight is strongly associated with
socioeconomic outcomes later in life (Black, et al. 2007; Behrman and Rosenzweig, 2004).
These links allow us to formulate a chain of effects that connect high levels of inequality with pre-natal
maternal stress, which in turn has the effect of reducing birth weight, a significant determinant of
malnutrition during infancy. The effects of stress on health through fetal growth are important, not only
because of the increased risk of malnutrition but also because of the long term effects that LBW has on
adult health. Couzin (2002) summarizes how endocrinologist Hobathan Seckl, of Western General
Hospital in Edinburgh, U.K., believes that excess levels of stress hormones in the fetus “reset” an
important arbitrator of stress in the body, making it hypersensitive to even banal events. In other words,
the body secretes glucose, cortisol, and other stress-related elements when they wouldn’t be normally
needed – a result now observed in LBW humans. Studies in humans and animals suggest that individuals
who are under the growth curve and need to catch-up to the rest are in a situation where adult illnesses
linked with LBW are more probable (Couzin, 2002). Currie and Hyson (1999) measure how the effects of
LBW persist well into adulthood and find that the effects are greatest for educational attainment, followed
by self-reported health status and employment (Currie & Hyson, 1999). Behrman and Rosenzweig (2004)
offer estimates which show that intrauterine nutrient intake significantly affects adult height (Behrman &
Rosenzweig, 2004). Almond et al. (2005) found that LBW infants tend to have lower educational
attainment, poorer self-reported health status, and reduced employment and earnings as adults, relative to
their normal weight counterparts (Almond, et al., 2005). If there is a connection between pre-natal stress
and health outcomes in adulthood, then inequality may be an important determinant of those outcomes in
11
countries where it is high enough to produce high levels of stress among the most disadvantaged and
specifically among disadvantaged women who are pregnant.
In the following graph by Holzman et al. (2001) we can see the various ways in which social context and
stress may be conducive to a preterm delivery where babies are generally underweight. In this layout
stress not only affects CRH levels but also may increase infection and vascular disease.
Source: C. Holzman, et al., 2001, Pregnancy outcomes and community health: the POUCH study of preterm delivery, Paediatric and perinatal Epidemiology, 15(2), pp. 138.
b. Inequality and maternal agency over household resources
The effect on inequality on malnutrition, in this pathway, is executed through the eroding effect that
inequality has on social cohesion. We argue inequality is a measure of social cohesion because the
evidence suggests that more egalitarian societies tend to be more cohesive (Wilkinson, 2000; Kawachi, et
12
al., 1999; Wilkinson, 1996), less violent (Wilkinson, 2000; Hsieh & Pugh, 1993), and hostile (Wilkinson,
2000; Williams, et al., 1995; Wilkinson, 1999), more trustworthy (Wilkinson, 2000; Kawachi, et al.,
1997) and have a higher levels of community life participation (Wilkinson, 2000; Putnam, et al., 1993). In
this sense, we are defining social cohesion as a set of social networks and capital which form when there
is a harmonious economic and social development of a community, where there is equal access to
services and welfare benefits, where there is a high degree of interaction within the community and
families, where there is a smooth resolution of collective action problems (Forest & Kearns, 2001).
Speer et al. (2001) provide evidence that social cohesion is related to interpersonal empowerment as
defined by a person’s leadership abilities, tendency to take a leadership role, and preference for leadership
positions. Peterson and Hughey (2004) find that there are differentiated effects of social cohesion on
interpersonal empowerment for men and women. The authors find that women need to invest in a
network of complex relationships before participation can help them feel empowered. We propose that
higher levels of social cohesion may in fact foster the creation of these networks and be conducive to
women’s interpersonal empowerment at the community level.
We argue that women who experience an erosion of their interpersonal empowerment at the community
level may also experience an erosion of their agency over household resources. This lack of agency may
have the effect of refocusing these resources away from the needs of children. There is recent evidence of
an association between women’s empowerment and the health of their children arising mainly from Asia.
Maitra (2004) find that women having control over household resources, such as the ability to keep
money, has a significant effect on the demand for prenatal care and hospital delivery in India. Shroff et al.
(2009) find that women who have access to discretionary money and permission to go to the market
(financial and physical autonomy) were significantly associated with lower stunting of children also in
India. They propose that the autonomy of the mother allows her to gain control of, and access to resources
which make it more likely she will provide effective child nutrition. More recently, Imai, et al. (2014)
find strong associations between women’s empowerment and the nutritional status of children in rural
India in the long run at the low end of its conditional distribution.
This evidence demonstrates that the same income level in a household can lead to different decisions
depending on the balance of power between the parents (Basu, 2006). Some theoretical arguments on how
women’s status in the household affect a child’s nutritional status rests on an intra-household bargaining
model between the two parents where the mother, m, and the father, f, derive her utility, 𝑈𝑚(𝑥𝑚 , 𝑞; 𝐴𝑚),
and his utility, 𝑈𝑓(𝑥𝑓 , 𝑞; 𝐴𝑓), from own consumption of commodities 𝑥𝑖 and child health, q. Children a
considered to be a “public good” for both parents and are not decision-makers (Basu, 2006).
13
𝛾𝑈𝑚(𝑥𝑚 , 𝑞; 𝐴𝑚) + (1 − 𝛾)𝑈𝑓(𝑥𝑓 , 𝑞; 𝐴𝑓) (1)
𝐼 = 𝑝𝑚𝑥𝑚 + 𝑝𝑓𝑥𝑓 + 𝑝𝑐𝑞, (2)
In this setting household utility (1) includes the bargaining power of the mother, γ, which ranges (0 <
𝛾 < 1) and 𝐴𝑖 which are exogenous household factors such as individual attitudes towards healthcare.
Here the mother and father are therefore assumed to make decisions about the quality of health of
children or spending on healthcare independently as part of each person’s utility function (Fafchamps, et
al., 2009; Maitra, 2004; Park, 2007; Basu, 2006).
By solving for the maximization of the household utility function subject to a budget constraint (2), where
𝑝𝑖 is the price of commodities for each parent and 𝑝𝑐 is the shadow price of public goods, that is, the
children, we obtain 𝑞∗ the health quality of the child which would depend on the following (Basu, 2006;
Fafchamps, et al., 2009; Maitra, 2004; Park, 2007):
𝑞∗ = 𝑞∗(𝛾, 𝐼, 𝑝𝑚 , 𝑝𝑓 , 𝑝𝑐 , 𝐴𝑚 , 𝐴𝑓) (3)
If we assume the mother is more likely than the father to value q, a higher γ or bargaining power of the
mother would result in better nutrition for the child. Additionally, we see that an increase in income, I,
also improves the health of the child (Imai, et al., 2014; McElroy, 1990; McElroy & Horney, 1981; Basu,
2006).
In conclusion, this pathway argues that social cohesion is protective of the interpersonal and economic
empowerment of mothers which increases their agency over household resources. This is conducive to a
greater focus on children’s nutritional health because we assume, and there is evidence, that mothers with
more financial and personal autonomy at the household level increases the focus of household resources
towards the needs of the children. Therefore, an erosion of social cohesion may also erode a mothers
agency over household resources and her capacity to property care and nurture her child.
3. Data
There are three phases of data processing. The first phase consists of calculating the z-score of height for
age using anthropometric data from the LSMS (2006). In the second phase we calculate per-capita
household consumption, the Gini coefficient and the incidence of poverty for every province, county and
parish. This is obtained using small area estimates which combines two data sources: the LSMS (2006)
and the Ecuadorian population census of 2010. Finally, the third phase consists of running OLS,
multilevel and logistical multilevel regressions on the LSMS (2006) where we use the 𝑧-score of height
for age and the dummy of being chronically malnourished as the dependent variable and the estimated
14
Gini coefficient as the independent variables of interest. In this section, we will go over the data and in
the next section we will go over the methodology used to estimate our results.
i. The Living Standards Measurement Survey of 2006
The LSMS of 2006 has national coverage, 55666 observations over 16414 households. Ecuador is divided
into 25 provinces, 224 counties and 1024 parishes; the LSMS covers 22 provinces, 186 counties and 443
parishes. The questionnaire goes over various topics such as living conditions, education, health care,
employment, food consumption and non-food consumption, income, access to credit, migration, and
economic activities such as entrepreneurship and agriculture. This survey includes anthropometric
measures for 6003 children under the age of five (Larrea & Kawachi, 2005). In table 1 we have
descriptive statistics for the variables which are later used in our model, as we can see the incidence of
chronic child malnutrition is 25% of the total population. All, except for the Gini coefficients, are taken
directly from the LSMS 2006.
15
Table 1 Descriptive statistics Variable Obs Weight Mean Std. Dev. Min Max
z-score 6003 1425121 -1.19 1.20 -5.5 4.1 dummy stunting 6003 1425121 0.25 0.44 0 1 Ln(consumption per capita) 5986 1422087 4.05 0.72 0.8 6.5 Gini province* 6003 1425121 0.40 0.05 0.3 0.6 Gini county* 6003 1425121 0.39 0.04 0.3 0.6
Gini parish* 6003 1425121 0.38 0.04 0.3 0.6 Ln(GDP province) 5840 1356020 21.94 1.32 18.7 23.3 Ln(GDP county) 5825 1352928 20.16 2.19 15.6 23.1 Incidence of poverty province 6003 1425121 0.40 0.20 0.1 0.8 Incidence of poverty county 6003 1425121 0.40 0.22 0.1 0.9
Incidence of poverty parish 6003 1425121 0.40 0.22 0.1 1.0 Mean consumption parish 6003 1425121 114.27 48.38 20.0 254.1 Mean consumption county 6003 1425121 114.61 47.39 25.4 203.5 Mean consumption province 6003 1425121 114.27 48.38 20.0 254.1 Dummy female 6003 1425121 0.50 0.50 0 1
Proportion of required vaccines 6003 1425121 0.75 0.23 0 1 Number of months of breastfeeding 5638 1345496 4.35 2.56 0.0 24.0 Dummy for low birth weight 6003 1425121 0.03 0.18 0 1 Years of schooling of the mother 5909 1401471 8.48 4.13 0.0 22.0 Age of the mothers 5940 1409448 28.13 7.29 12.0 64.0
Fertilidad de la madre 5847 1387869 0.16 0.08 0.0 0.7 Dummy birth with cesarian section 6003 1425121 0.36 0.48 0 1 Dummy birth with Phisysian or Obstetritian 6003 1425121 0.80 0.40 0 1 Dummy mother underemployed 6003 1425121 0.35 0.48 0 1 Index of housing conditions 6003 1425121 -0.18 1.00 -2.5 1.2
Years of experience of the head of the household 5984 1420927 24.19 14.44 0.0 55.0 Square root of number of children in household 6003 1425121 1.55 0.43 1.0 3.2 Dummy government cash transfer 6003 1425121 0.37 0.48 0 1 Dummy government nutritional supplement 6003 1425121 0.27 0.45 0 1 Rate of assistance to post-secondary education 6003 1425121 0.14 0.20 0 1
Gini of Land 5975 1421535 0.76 0.11 0.4 0.9 Index of food consumption 5850 1404743 -0.43 1.46 -4.9 7.0 Dummy rural 6003 1425121 0.40 0.49 0 1 Number of doctors per every 10.000 people 5999 1424070 6.50 16.60 0.0 130.4 Dummy amazonia rural 6003 1425121 0.05 0.21 0 1
Categories of ethnicities
0 - Mestizo 4,508 1146396 80.44
1 - Indigenous 1,073 189198 13.28
2 - Afro-ecuadorian 422 89527 6.28
Source: Small Area Estimates using Living Standards Measurement Survey, Ecuadro 2006 (Encuesta de Condiciones de Vida, 2006) and
Ecuadorian Census of 2010. Instituto Nacional de Encuestas y Censos, Ecuador. Data procesing: authors with Unidad de Infromación Socio
Ambiental, Universidad Andina Simón Bolívar.
*It is not possible to take the average of a Gini coefficient given it is not decomposable. This figure is not the Gini coefficient of the country it is here simply to provide the reader an idea of the range of the data.
ii. Dependent variable: Chronic child malnutrition
We estimate the 𝑧-score of height for age using the methodology developed and distributed freely by the
World Health Organization (2013). The normalized 𝑧-score establishes the growth standard of children by
defining a normal growth curve and is defined as follows (World Health Organization, 2013; World
Health Organization, 1997).
𝑧 𝑠𝑐𝑜𝑟𝑒 =(𝑥𝑖 − 𝑥𝑚𝑒𝑑𝑖𝑎𝑛)
𝜎𝑥⁄
16
Where 𝑥𝑖 is the height of child i, 𝑥𝑚𝑒𝑑𝑖𝑎𝑛 is the median height from the reference population of the same
age and gender and 𝜎𝑥 is the standard deviation of 𝑥 of the same reference population (Imai, et al., 2014;
World Health Organization, 1997). We use anthropometric data available in the LSMS (2006) to calculate
the normalized 𝑧-score for each child below the age of five. The 𝑧-score ranges from −∞ to ∞ as it is
measured in standard deviations from the mean which is zero. If a child’s 𝑧-score is under -2, that is to
say, under two standard deviations below the mean, the child is chronically malnourished or “stunted”
(World Health Organization, 1997). Graph 1 and 2 shows the 𝑧-score average in every sub-region and its
distribution, and Map 1 shows the percentage of children with chronic malnutrition for every parish
(calculated using small area estimates explained below).
Graph 1 Box plot z-score over sub-regions
Source: Living Standards Measurement Survey, Ecuadro 2006 (Encuesta de Condiciones de Vida, 2006). Data procesing: authors with Unidad
de Infromación Socio Ambiental, Universidad Andina Simón Bolívar.
17
Graph 2 Z-score distribution in Ecuador
Source: Living Standards Measurement Survey, Ecuadro 2006 (Encuesta de Condiciones de Vida, 2006) Instituto Nacional de Encuestas y
Censos, Ecuador. Data procesing: authors.
As we can see, the national average z-score is -1.217 and, in the rural highlands the average is close to -2.
This demonstrates a deteriorated nutritional state among children and a concentration of chronic
malnutrition in the rural highlands which we can see in Map 1.
Map 1 Incidence of chronic child malnutrition in Ecuador by parish
Source: Small Area Estimates using Living Standards Measurement Survey, Ecuadro 2006 (Encuesta de Condiciones de Vida, 2006) and
Ecuadorian Census of 2010. Instituto Nacional de Encuestas y Censos, Ecuador. Data procesing: authors with Unidad de Infromación Socio
Ambiental, Universidad Andina Simón Bolívar.
18
iii. Independent variable: The Gini Coefficient
The Gini coefficient is measured at the parish, county and provincial level using small area estimates. We
chose to use this methodology because household surveys that include measures of income and
consumption are rarely representative at the local level or of sufficient size to yield statistically reliable
estimates of poverty or inequality. Similarly, census data, which is generally of sufficient size to allow for
reliability at low levels of aggregation (small geographic scales), usually do not have information about
income or consumption, as is Ecuador’s particular case. Elbers, et al. (2003) estimate consumption,
poverty and the Gini coefficient using this method on Ecuadorian data. They report that their results have
levels of precision comparable to those of survey based estimates, but for populations as small as 15000
households. The combination of the census data and the survey data allows for estimations of the Gini
coefficient in subpopulations one hundredth the size of the subpopulations in the survey data and yet
obtain very similar prediction errors (Elbers, et al., 2003). For a detailed explanation of the methodology
refer to Appendix 1.
This methodology has not gone without critique. Tarozzi & Deaton (2009) argue that, in order to match
survey and census data in the way which is proposed by Elbers et al. (2003), a degree of spatial
homogeneity is required for which the method has no basis. They argue that estimates based on those
assumptions may underestimate the variance of the error in predicting welfare estimated at the local level
and therefore overstate the coverage of confidence intervals (Tarozzi & Deaton, 2009). In response,
Elbers, et al. (2008) compare their small area estimate welfare results in Minas Gerais, Brasil, a notably
heterogeneity area, with the true welfare values and find that the methodology yielded welfare estimations
which were close to these true values and had confidence interval estimations which were appropriate.
This demonstrates that if the methodology is applied with careful control over the conditional distribution
of income, the estimations can be reliable (Elbers et al., 2008).
In this case, as is the case Elbers, et al. (2003) (see Appendix 1 for comparative results), Ecuador is
divided into eight sub-regions. Firstly three general geographic regions: coast, highlands and amazon
basin which are divided into rural and urban areas excluding the two largest cities which are the final two
region-cities: Quito and Guayaquil. We run a separate consumption model for each of these sub-regions,
and therefore we predict household consumption in eight separate simulation models depending on where
the household is located.With these household consumption simulation results we estimate Gini
coefficients over every province, county and parish of the population census. Given there can be both
rural and urban areas within the same province, county or parish, separating them increases the level of
19
homogeneity within each sub-region and within each small area estimate model we run (see appendix 2
for detailed on consumption models per sub-region).
We present our results in Table 4 (and our extended results in appendix 1) and compare them to the Gini
coefficients estimated directly from the LSMS (2006) over the sub-regions. As shown, there is a degree of
discrepancy between the LSMS and the SAE point estimations. This may due to the fact that the models
tend to underestimate the household consumption of high income homes given that the variables we use
in our equation (Appendix 2) estimations are generally measuring lack of resources.
Table 2 Comparison of estimations of the Gini coefficient from the LSMS (2006) and using SAE (2006-2010)
Region
Regional Gini
LSMS (2006) SAE (2006-2010)
Quito 0.463 0.403 Guayaquil 0.416 0.386 Urban Coast 0.409 0.358 Rural Coast 0.357 0.281 Urban Highlands 0.411 0.346 Rural Highlands 0.454 0.387 Urban Amazon 0.416 0.355 Rural Amazon 0.47 0.454 National Total 0.466 0.419
Source: Small Area Estimates using Living Standards Measurement Survey, Ecuadro 2006 (Encuesta de Condiciones de Vida, 2006) and
Ecuadorian Census of 2010. Instituto Nacional de Encuestas y Censos, Ecuador. Data procesing: authors with Unidad de Infromación Socio
Ambiental, Universidad Andina Simón Bolívar.
The results of our estimations of small areas can be seen in Map 2.a and compared to those estimated by
the government of Ecuador presented in Map 2.b. The latter shows the results obtained by the Ministry of
Social Development of Ecuador which published these results in January of 2012. They also follow the
same small area estimate methodology, however they decide to run a different model for every province
irrespective or rural or urban areas or of the influence of larger cities. Notwithstanding, their results at the
provincial level, the only results which they publish in the form of a map in their report, are similar in the
way in which the geographic distribution of the Gini coefficient is spread across the country. In this sense,
we see relatively higher Gini coefficients in the Amazon basin on the far right side of the map, relatively
lower coefficients in the coastal region, on the far left and relatively intermediate levels running down the
middle Andean corridor. This demonstrate a certain level of comparability of our results with official
Ecuadorian government data as well as an acceptance of the methodology as an appropriate method in the
study of social phenomenon and as a tool for public policy (Unidad de Análisis de Información del
Ministerio de Coordinación de Desarrollo Social, 2012).
20
Map 2 Gini coefficient in Ecuador
Map 2.a. This study’s results at parish level Map 2.b. Official government publication at province level
Map 2.a. Source: Small Area Estimates using Living Standards Measurement Survey, Ecuadro 2006 (Encuesta de Condiciones de Vida, 2006)
and Ecuadorian Census of 2010. Instituto Nacional de Encuestas y Censos, Ecuador. Data procesing: authors with Unidad de Infromación Socio
Ambiental, Universidad Andina Simón Bolívar. Map 2.b. Source: Small Area Estimates using Living Standards Measurement Survey, Ecuador
2006 (Encuesta de Condiciones de Vida, 2006) and Ecuadorian Census of 2010. Instituto Nacional de Encuestas y Censos, Ecuador. Data
procesing: Unidad de Análisis e Infromación de la Secretaría Tecnica del Ministerio de Coordinacion del Desarrollo Social SIISE STMCDS.
iv. Other regressors
The control variables include what the previous literature on the social determinants of malnutrition has
shown influential (Larrea, Freire, & Lutter, 2001; Larrea, 2002; Marins & Almeida, 2002; Willey et. al.,
2009; Aerts, et al., 2004; El Taguri, et al., 2009; Adair & David, 1997). These variables are measured
either at the individual level or are aggregated at a geographic level and can be grouped into 4 categories:
the characteristics of child, the characteristics of the mother, the characteristics of the household, and, the
characteristics of the greater context. This last one is defined at various levels of aggregation depending
on the nature of the information that is available.
In the first category we are attempting to control for the individual level characteristic which may affect
the z-score of the child. In order to do this we include age of the child (in months), the gender of the child,
the number of months of breastfeeding the child received, the proportion of vaccines the child has
received form those which are mandatory in early childhood,8 a dummy variable which discerns if the
child was born with a “low” birth weight. Low birthweight is typically defined as children weighing less
than 2500g. In some cases, the mothers did not provide the exact weight of the child at birth; however, the
survey does ask whenever the mother was told by her doctor or practitioner that the child was
“underweight.” This variable was included to visualize the effect of the first pathway in which we argue
8 BCG for tuberculosis, Pentavalente which is DTP for diphtheria, tetanus and pertussis, Hb for hepatitis B and HIB for type b
Haemophilus influenzae, poliomyelitis, and finally, measles.
21
the stress levels of the mother caused by inequality could lead to LBW which is a known determinant of
stunting.
Given that the mother is generally the main provider of care in early childhood, as well as carrying the
child over his or her gestation we control for the characteristics of the pre-natal, natal and post-natal
health services the mother has access to. we also include the number of years of education obtained, as
well as the age and fertility of the mother. 9
Variables regarding the mother’s vaccines (tetanus) and the
number of visits in the first and second trimester were not significant and were not included. However, a
dummy variable discerning whether there was a physician or obstetrician present while giving birth, and a
dummy variable discerning if she had a caesarian section were significant and included. It is not clear
why the second variable is significant; we suspect it captures access to health services which are not
widely available rather than the effect of a caesarian section per se. We also tested variables of the
mother’s employment10
and found that the dummy variable of whether the mother is under-employed11
was significant. We believe this variable is relevant mainly because it often shows a precarious working
environment, particularly when the mother falls into the second category of underemployment (see
footnote 9). Given that mothers tend to bring their young children to work, this may also affect their
health.
The characteristics of the household include variables which may affect the child during the pregnancy or
after birth. We control for the log of the per capita household consumption. This variable is measured in
log form so as to capture the concave effect that income has on health. We also include controls for the
square root of the number of children under the age of twelve living in the household and the years of
work experience of the head of the household. The first variable allows us to control for the number of
children in households where many generations live together. The second variable allows us to capture
the level of employment stability of the head of the household. We suspect this may have an effect on
access to better living arrangements beyond what income or household conditions may be able to provide.
For example, employment stability may provide a less stressful environment which may reduce the level
of domestic violence against women or children. Domestic violence indicators are not available in the
data therefore variables such as this one are the only indicators of low stress household environments.
The final household level variable, which measures housing conditions, is obtained using principal
components analysis. This methodology allows us to avoid using housing variables which may be highly
9
The number of live births a woman has had is divided by her age since adolescence. 10
Industry, unemployment and working conditions. 11
We defined underemployed as a person who falls into one of any three categories. Firstly, an employed person who works less
than 40 hours a week and is willing to work more hours, an employed person who makes less than $156.21 per month and works more than 40 hours a week or an employed person who works less than 40 hours a week however makes less than $156.21/h of work.
22
correlated in our regression model in order to circumvent high levels of multicollinearity. The index gives
high scores to households with access to basic services such as potable water, sewage connection,
electricity, public garbage collection services, landline telephone services; households which have
suitable constructions materials for walls, floors and roofs; finally, households which have exclusive use
of a washroom and are not overcrowded (see Appendix 4 for detailed results).
Finally, the contextual variables, apart from our variable of interest which we have explained above,
include firstly, the head count for poverty in the geographic area of interest.12
We include this variable
because of the effect that poverty may have on neighborhood environments or on the social fabric of
cities. This is an aggregated variable therefore it does not measure the same phenomenon as the
household per capita consumption. We also include the mean consumption of the specified geographical
area13
as a contextual variable given there is some evidence that having wealthier neighbors may create
positive externalities in access to healthcare and individual health (Miller & Paxson, 2006). Additionally,
we include the GDP of the specified geographical area,14
in order to capture the effect of aggregate
production on individual health, given we believe that living in areas with greater levels of production
may increase the access to specific health and welfare services. We also include the rate of attendance to
higher education for similar reasons; we believe that having a more educated community has a positive
externality on the health and child care behaviours of individuals living in that community. Additionally,
we include the Gini of land in order to control for agricultural production conditions of the rural areas.
Finally, we include a food consumption index which we explain below.
The food consumption index identifies the average carbohydrates, fat and protein consumed in each
parish (Programa Alimentate Ecuador, Ministerio de Inclusión Económica y Social, Ecuador, 2009). The
index gives high scores to parishes with high carbohydrate consumption, indicating a severe protein
deficiency and a lack of micronutrient intake from fruits15
(see Appendix 5 for details) (Larrea &
Kawachi, 2005). This index gives us an idea of the social and institutional barriers present in the access to
proper nourishment and food security.
The rural Amazon can be characterized as fairly heterogeneous sub-region. It is a sub-region with
considerable historic and economic differences to the rest of the country. For example, this the home of
various relatively isolated first nations or indigenous groups such as the Waorani. The Waorani are highly
mobile semi-nomadic, hunter-gatherer-horticulturalists who live in various widely dispersed groups
(Finer, et al., 2009). Additionally, this sub-region is where extensive oil extraction activities and
12 Be it province, county or parish. 13 Province, county or parish, depending on the model. 14
Province, county or parish, depending on the model. 15
Essentially a potato based diet
23
important environmental impacts of these activities take place, unlike in the rest of the country (Finer, et
al., 2009). The particular lifestyles of these groups along with the oil extraction activities may have an
impact on human health which we cannot control for. Therefore, we decided to include a rural amazon
dummy. This variable relates to the geographic divisions of provinces in what is known as a cross
classified form. Therefore individuals living in the rural amazon can live in any province in the amazon
and all provinces in the amazon have rural areas. In this sense, this dummy variable is not nested within a
province but rather is cross classified through all provinces in the amazon and vice versa. Therefore this
dummy variable is not orthogonal in any way to the provinces which are included in the multilevel model
as a level.
v. Fixed effects
We use fixed effects to control for the effect of ethnicity. Given this is a cross-section analysis, fixed
effects amount to a dummy variable for each category of the following three categories: mestizo16
,
indigenous17
and afro-Ecuadorian (Larrea & Kawachi, 2005). Given the multilevel model controls for the
heterogeneity within each area (province) we decided not to add additional regional fixed effects beyond
a dummy variable for rural areas.
4. Methodology
Multilevel modeling techniques are generally used in this type of analyses given they allow us to account
for clustering within communities and for the examination of both inter-individual and inter-group
variation, as well as the interactions between levels (VanderEnde, et al., 2012; Diez-Roux, 2000;
Raudenbush & Bryk, 2002). These types of regressions are said to enhance the scope of the OLS models
by allowing for the decomposition of the residual variance into contextual and individual factors. The
multilevel models have an individual fixed component and a random component which allows for the
modelling of contextual heterogeneity (Larrea & Kawachi, 2005; Snijders & Bosker, 1999).
We report results from three types of models, the OLS estimations, the multilevel estimations and the
logistical multilevel estimations. For the first two models, we use the z score of height-for-age as the
dependent variable. For the last model, we use a dummy variable for chronic malnutrition. This variable
is therefore equal to 1 if the z-score of the child is below -2 or equal to 0 if the z-score is -2 or higher.
16
Defined as a mix between Caucasian and Indigenous. 17 Indigenous people are usually fluent in one of various indigenous languages as well as Spanish, whereas it is rare for non-
indigenous peoples to be fluent in such languages. Therefore it may be suggested that the fluency in these indigenous languages be used as a proxy for ethnicity. Specifically, we find an individual to be indigenous if she or he either states it directly, or if she or he states that the spoken language in the household is an indigenous one. We do this in order to avoid an underestimation of the indigenous population.
24
We present two types of multilevel models: two random slope models and one random intercept model.
Both models allow us to have a fixed component and a random component. However, the random
intercept model allows the intercept of the equation to vary within each province and the random slope
model allows both the intercept and the slope of the relationship to vary within each province. In our
models we have individuals nested in parishes which are nested in provinces. Our dependent variable is
measured at the individual level, our independent variable, the Gini coefficient, is measured at the parish
or county level, and we allow this slope and the intercept to vary within each province. Therefore, when
we measure the Gini coefficient at the parish level we build a random slope model where both the
intercept and the slope of the inequality nutrition relation may vary within each province. With this model
we are allowing for the relation to vary within provinces because we think that the behaviour of the
outcome variable may be similar in parishes which are in the same provinces and differ across parishes
which are nested in different provinces. We do the same when we measure the Gini coefficient at the
county level.
When measuring the Gini coefficient for provinces, we can no longer have a random slope model, given
there is only one estimation of this coefficient for every province. Therefore, in this case, we use a
random intercept model where the intercept is allowed to change within each province however the slope
is part of the fixed section of the model. We repeat the same logic for the logistical multilevel models.
Here below we present the equation of our estimations.
OLS with fixed effects for ethnicity model:
𝑦𝑖𝑗 = 𝛽0 + 𝛽1𝐺𝑖𝑛𝑖𝑗 + 𝑋𝛽 + 𝜂𝑖 + 𝑒𝑖𝑗
𝑒𝑖𝑗~𝑁(0, 𝜎2)
Multilevel models:
Random slope models for parish and county: Here we have a relationship between the z-score of the child
and the parish/county Gini coefficient which is allowed to vary in slope and intercept depending on which
province the parish/county is located in.
𝑦𝑖𝑗𝑘 = 𝛽0 + 𝛽1𝐺𝑖𝑛𝑖𝑗𝑘 + 𝜂𝑖 + 𝑢0𝑘 + 𝑢1𝑘𝐺𝑖𝑛𝑖𝑗𝑘 + 𝑒0𝑖𝑗
𝑒0𝑖𝑗𝑘~𝑁(0, 𝜎𝑒02 )
[𝑢0𝑘
𝑢1𝑘] ~𝑁(0, Ω𝑢)
Ω𝑢 = [𝜎𝑢0
2
𝜎𝑢01 𝜎𝑢12 ]
25
Ginijk is the Gini coefficient of parish/county j in province k.
u0k is the random intercept of province k.
u1k is the random slope of province k.
Random intercept models for province: Given the Gini of province cannot have a random slope as it is not
nested in any greater area we perform a random intercept model, where only the intercept of every
province is allowed to vary.
𝑦𝑖𝑘 = 𝛽0 + 𝛽1𝐺𝑖𝑛𝑖𝑘 + 𝜂𝑖 + 𝑢𝑘 + 𝑒𝑖𝑘
𝑒𝑖𝑗~𝑁(0, 𝜎𝑒2)
𝑢𝑖𝑗~𝑁(0, 𝜎𝑢2)
Ginijk is the Gini coefficient of province k.
uk is the random intercept of province k.
Logistical multilevel models
Random slope models for parish and county: Again, here we have a relationship between the z-score and
the Gini coefficient at the parish/county level which varies according to the province in which the
parish/county is in.
log[𝑝𝑖𝑗𝑘/(1 − 𝑝𝑖𝑗𝑘)] =𝛽0 + 𝛽1𝐺𝑖𝑛𝑖𝑗𝑘 + 𝑢1𝑘𝐺𝑖𝑛𝑖𝑗𝑘 + 𝑢0𝑘 + 𝑒0𝑘
𝑒0𝑖𝑗𝑘~𝑁(0, 𝜎𝑒02 )
[𝑢0𝑘
𝑢1𝑘] ~𝑁(0, Ω𝑢)
Ω𝑢 = [𝜎𝑢0
2
𝜎𝑢01 𝜎𝑢12 ]
Random intercept models for province: Here we cannot build a model with a random slope for province
either given there is only one estimation of Gini of province for each province. Therefore here, as above
26
we perform a random intercept model, where only the intercept of each province is allowed to change
over provinces.
log[𝑝𝑖𝑗𝑘/(1 − 𝑝𝑖𝑗𝑘)] =𝛽0 + 𝛽1𝐺𝑖𝑛𝑖𝑗𝑘 + 𝑣0𝑘 + 𝑢0𝑘
𝑒𝑖𝑗𝑘~𝑁(0, 𝜎𝑒2)
𝑢𝑘~𝑁(0, 𝜎𝑢2)
5. Main Findings
In the following section we present out estimation results and discuss the main findings. The results are
organized in terms of the type of model, firstly we present the OLS regressions, then the multilevel
regression and finally, the logistic multilevel regressions with both raw and marginal effects.
27
Table 3 OLS Regression Models
OLS1
OLS2
OLS3
b/se
b/se
b/se
Gini province -1.565* (0.63) Gini county
-1.579* (0.63)
Gini parish
-1.481* (0.61) Ln(H. Consumption) 0.148*** (0.04) 0.150*** (0.04) 0.150*** (0.04) Ln(GDP) 0.126*** (0.02) 0.056*** (0.01) 0.049*** (0.01) Poverty 0.658* (0.30) -0.619 (0.37) 0.099 (0.48) Mean consumption -0.002* (0.00) -0.006*** (0.00) -0.004* (0.00) Characteristics of the child
D. female 0.146*** (0.03) 0.154*** (0.03) 0.155*** (0.03) Age of child in months -0.140*** (0.02) -0.137*** (0.02) -0.137*** (0.02) Age of child in months2 0.004*** (0.00) 0.004*** (0.00) 0.004*** (0.00) Age of child in months3 -0.000*** (0.00) -0.000*** (0.00) -0.000*** (0.00) Proportion of vaccines received 0.139 (0.10) 0.083 (0.10) 0.077 (0.10) N. months of breastfeeding -0.004 (0.01) -0.005 (0.01) -0.006 (0.01) Low birth weight -0.596*** (0.10) -0.592*** (0.10) -0.594*** (0.10) Characteristics of the mother
Schooling of the mother 0.028*** (0.01) 0.027*** (0.01) 0.027*** (0.01) Age of the mother 0.010*** (0.00) 0.010*** (0.00) 0.010*** (0.00) Fertility of mother -0.924** (0.31) -0.946** (0.32) -0.935** (0.32) D. caesarian section 0.119** (0.04) 0.142*** (0.04) 0.142*** (0.04) D. obstetrician/physician birth 0.068 (0.05) 0.073 (0.05) 0.087 (0.05) D. mother underemployed -0.066 (0.03) -0.067 (0.04) -0.067 (0.04) Characteristics of the household
Housing conditions index 0.056 (0.03) 0.029 (0.03) 0.033 (0.03) Work experience of head of household 0.002 (0.00) 0.002 (0.00) 0.002 (0.00) N. children < 12 in household 1/2 -0.159** (0.06) -0.169** (0.06) -0.172** (0.06) Cash transfer -0.081* (0.04) -0.068 (0.04) -0.071 (0.04) Nutritional supplement -0.069 (0.04) -0.062 (0.04) -0.064 (0.04) Other contextual variables
R.A. higher education census sector 0.305** (0.10) 0.294** (0.10) 0.252** (0.10) Gini Land -0.529** (0.19) -0.424* (0.20) -0.450* (0.20) Food consumption index -0.081*** (0.01) -0.105*** (0.01) -0.109*** (0.01) D. rural -0.335*** (0.08) -0.195** (0.07) -0.282*** (0.08) Number public doctors/10000 ppl. 0.001 (0.00) 0.001 (0.00) 0.001 (0.00) Dummy rural amazon 0.424*** (0.12) 0.376** (0.11) 0.292** (0.11) Fixed effects for ethnicity
Indigenous -0.195*** (0.06) -0.176** (0.06) -0.180** (0.06) Afro-Ecuadorian 0.236** (0.07) 0.228** (0.07) 0.213** (0.07) _cons -2.419*** (0.48) 0.049 (0.43) -0.361 (0.47)
r2 0.275
0.267 5103 0.266 5103 N 5117
5103
5103
(+p< 0.10, *p< 0.05, ** p<0.01, *** p<0.001)
28
Table 4 Multilevel Regression Models
Random intercept Random slope Random slope
(1) (2) (3)
b/se
b/se
b/se
Gini province -2.173* (0.94) Gini county
-1.769*** (0.53)
Gini parish
-1.223* (0.52) Ln(H. Consumption) 0.140*** (0.03) 0.142*** (0.03) 0.141*** (0.03) Ln(GDP) 0.076** (0.03) 0.012 (0.01) 0.005 (0.01) Poverty 0.437 (0.58) 0.054 (0.54) -0.190 (0.41) Mean consumption -0.002 (0.00) -0.003 (0.00) -0.003 (0.00) Characteristics of the child
D. female 0.160*** (0.02) 0.160*** (0.02) 0.160*** (0.02) Age of child in months -0.149*** (0.02) -0.148*** (0.02) -0.148*** (0.02) Age of child in months2 0.004*** (0.00) 0.004*** (0.00) 0.004*** (0.00) Age of child in months3 -0.000*** (0.00) -0.000*** (0.00) -0.000*** (0.00) Proportion of vaccines received 0.154* (0.08) 0.149 (0.08) 0.146 (0.08) N. months of breastfeeding -0.003 (0.01) -0.003 (0.00) -0.003 (0.00) Low birth weight -0.603*** (0.06) -0.604*** (0.07) -0.603*** (0.06) Characteristics of the mother
Schooling of the mother 0.029*** (0.01) 0.028*** (0.01) 0.028*** (0.01) Age of the mother 0.010*** (0.00) 0.010*** (0.00) 0.010*** (0.00) Fertility of mother -0.699* (0.32) -0.697* (0.33) -0.699* (0.33) D. caesarian section 0.105** (0.04) 0.115** (0.04) 0.115** (0.04) D. obstetrician/physician birth 0.087 (0.06) 0.084 (0.06) 0.088 (0.05) D. mother underemployed -0.089** (0.03) -0.093** (0.03) -0.095** (0.03) Characteristics of the household
Housing conditions index 0.071* (0.03) 0.067* (0.03) 0.064* (0.03) Work experience of head of household 0.003 (0.00) 0.003* (0.00) 0.003* (0.00) N. children < 12 in household 1/2 -0.180*** (0.05) -0.187*** (0.05) -0.189*** (0.05) Cash transfer -0.080 (0.04) -0.072 (0.04) -0.078 (0.04) Nutritional supplement -0.097* (0.04) -0.095* (0.04) -0.094* (0.04) Other contextual variables
R.A. higher education census sector 0.264** (0.08) 0.309*** (0.09) 0.307*** (0.08) Gini Land -0.486 (0.27) -0.464 (0.29) -0.527 (0.30) Food consumption index -0.071** (0.02) -0.079*** (0.02) -0.081*** (0.02) D. rural -0.232** (0.08) -0.129* (0.06) 0.287* (0.14) Number public doctors/10000 ppl. 0.000 (0.00) 0.000 (0.00) -0.078 (0.06) Dummy rural amazon 0.441** (0.14) 0.352** (0.14) 0.001 (0.00) Fixed effects for ethnicity . . . . . .
Indigenous -0.159*** (0.04) -0.183*** (0.05) -0.182** (0.06) Afro-Ecuadorian 0.270*** (0.06) 0.278*** (0.06) 0.277*** (0.06) _cons -1.049 (0.78) 0.347 (0.58) 0.408 (0.53)
sd(GINI)
0.342 (0.0945) 0.313 (0.620) sd(_cons) 0.106 (0.0302) 0.0000108 (0.000686) 0.0679 (0.377) sd(Residual) 1.007 (0.0165) 1.007 (0.0390) 1.007 (0.0172)
N 5117 5103 5103
(+p< 0.10, *p< 0.05, ** p<0.01, *** p<0.001)
29
Table 5 Parrish Level Regression Models
Random intercept
Random slope
Random slope
(1)
(2)
(1)
(2)
(1)
(2)
raw
margin
raw
margin
raw
margin
Gini province 3.981** (1.406) 0.655** (0.23) Gini county
2.607+ (1.361) 0.44+ (0.23)
Gini parish
0.573 (1.284) 0.096 (0.21) Ln(H. Consumption) -0.302*** (0.0813) -0.049*** (0.01) -0.313*** (0.0817) -0.052*** (0.01) -0.325*** (0.0822) -0.054*** (0.01) Ln(GDP) -0.155** (0.0494) -0.025** (0.00) -0.0299 (0.0327) -0.005 (0.00) -0.0209 (0.0318) -0.003 (0.00) Poverty -1.228 (1.059) -0.202 (0.17) 0.332 (0.804) 0.056 (0.13) 1.220+ (0.691) 0.206+ (0.11) Mean consumption -0.000183 (0.00455) 0.00003 (0.00) 0.00411 (0.00393) 0.0006 (0.00) 0.00728* (0.00335) 0.001* (0.00) Characteristics of the child
D. female -0.314*** (0.0700) -0.051*** (0.01) -0.316*** (0.0701) -0.053*** (0.01) -0.322*** (0.0701) -0.054*** (0.01) Age of child in months 0.341*** (0.0360) 0.056*** (0.00) 0.339*** (0.0361) 0.057*** (0.00) 0.337*** (0.0361) 0.056*** (0.00) Age of child in months2 -0.00958*** (0.00118) -0.001*** (0.00) -0.00954*** (0.00118) -0.001*** (0.00) -0.00947*** (0.00118) -0.001*** (0.00) Age of child in months3 0.0000811*** (0.0000117) 0.00001*** (0.00) 0.0000808*** (0.0000117) 0.00001*** (0.00) 0.0000803*** (0.0000117) 0.00001*** (0.00) Proportion of vaccines received -0.294 (0.196) -0.048 (0.03) -0.268 (0.197) -0.045 (0.03) -0.251 (0.197) -0.042 (0.03) N. months of breastfeeding 0.000173 (0.0146) 0.00002 (0.00) 0.00295 (0.0147) 0.0004 (0.00) 0.00326 (0.0147) 0.0005 (0.00) Low birth weight 1.184*** (0.199) 0.195*** (0.03) 1.179*** (0.200) 0.199*** (0.03) 1.187*** (0.200) 0.200*** (0.03) Characteristics of the mother
Schooling of the mother -0.0538*** (0.0126) -0.008*** (0.00) -0.0521*** (0.0126) -0.008*** (0.00) -0.0513*** (0.0126) -0.008*** (0.00) Age of the mother -0.0228*** (0.00546) -0.003*** (0.00) -0.0221*** (0.00547) -0.003*** (0.00) -0.0223*** (0.00547) -0.003*** (0.00) Fertility of mother 2.054*** (0.623) 0.338*** (0.10) 2.027** (0.623) 0.342** (0.1) 1.999** (0.622) 0.338** (0.1) D. caesarian section -0.184* (0.0923) -0.03* (0.01) -0.213* (0.0925) -0.036* (0.01) -0.209* (0.0925) -0.035* (0.01) D. obstetrician/physician birth -0.260** (0.0963) -0.042** (0.01) -0.248* (0.0967) -0.041* (0.01) -0.238* (0.0977) -0.04* (0.01) D. mother underemployed 0.133+ (0.0722) 0.021+ (0.01) 0.145* (0.0723) 0.024* (0.01) 0.156* (0.0723) 0.026* (0.01) Characteristics of the household
Housing conditions index -0.0847 (0.0625) -0.013 (0.01) -0.0609 (0.0630) -0.01 (0.01) -0.0447 (0.0638) -0.007 (0.01) Work experience of head of household -0.00893** (0.00286) -0.001** (0.00) -0.00934** (0.00287) -0.001** (0.00) -0.00930** (0.00287) -0.001** (0.00) N. children < 12 in household 1/2 0.315** (0.119) 0.051** (0.01) 0.325** (0.119) 0.054** (0.02) 0.331** (0.119) 0.055** (0.02) Cash transfer 0.0996 (0.0817) 0.016 (0.01) 0.0945 (0.0820) 0.015 (0.01) 0.0999 (0.0820) 0.016 (0.01) Nutritional supplement 0.152+ (0.0849) 0.025+ (0.01) 0.151+ (0.0851) 0.025+ (0.01) 0.152+ (0.0852) 0.025+ (0.01) Other contextual variables
R.A. higher education census sector -0.392+ (0.230) -0.064+ (0.03) -0.450+ (0.232) -0.075+ (0.03) -0.472* (0.233) -0.079* (0.03) Gini Land 0.421 (0.452) 0.069 (0.07) 0.420 (0.503) 0.07 (0.08) 0.463 (0.507) 0.078 (0.08) Food consumption index 0.125*** (0.0319) 0.02*** (0.00) 0.143*** (0.0331) 0.024*** (0.00) 0.155*** (0.0331) 0.026*** (0.00) Number public doctors/10000 ppl. -0.00170 (0.00269) 0.0002 (0.00) -0.00144 (0.00270) 0.0002 (0.00) -0.00169 (0.00271) 0.0002 (0.00) Dummy rural amazon -0.729** (0.274) -0.12** (0.04) -0.583* (0.276) -0.098* (0.04) -0.341 (0.261) -0.057 (0.04) D. rural 0.315+ (0.182) 0.051+ (0.02) 0.0432 (0.159) 0.007 (0.02) -0.0445 (0.147) -0.007 (0.02) Fixed effects for ethnicity . .
. .
. .
Indigenous 0.227* (0.109) 0.039* (0.01) 0.262* (0.111) 0.046* (0.02) 0.254* (0.113) 0.045* (0.02) Afro-Ecuadorian -0.787*** (0.168) -0.115*** (0.02) -0.810*** (0.170) -0.12*** (0.02) -0.806*** (0.170) -0.12*** (0.02) _cons -0.301 (1.425)
-3.427 (1.077)
-3.487 (1.045)
sd(GINI)
0.577 (0.143) 0.577 (0.143) 0.594 (0.146) 0.594 (0.146) sd(_cons) 0.165 (0.0512) 0.165 (0.0512) 0.000119 (0.243) 0.000119 (0.243) 0.0000207 (0.433) 0.0000207 (0.433)
N 5117
5103
5103
(+p< 0.10, *p< 0.05, ** p<0.01, *** p<0.001)
30
i. The effect of inequality
The Gini coefficient is significant and has a deleterious effect in every OLS and multilevel model. In the
logistical multilevel models it is deleterious and significant when measured at the provincial level; it only
has a 10% significance level (p<0.1) when measured at the county level, and it is not significant when
measured at the perish level.
We can see that, in the OLS models, the magnitude of the effects are all quite similar (around -1.5) ,
while in the multilevel model the magnitude of the effect on the z-score when the Gini is measured at the
provincial level (-2.173) is larger than the effect when measured at the county level (-1.769) and is almost
twice as large of the effect when measured at the parish level (-1.223). Finally, in the logistical multilevel
model, the marginal effect of the Gini when measured at the provincial (0.655) is the largest and only
significant effect.
Larrea and Kawachi (2006) do no find a significant relationship between consumption inequality and
chronic child malnutrition at the county or at the parish level, in spite of using multilevel models in
addition to multivariable regressions. This may be due to the lack of quality data with information on
child nutrition, as well as the smaller sample size. The LSMS of 1998 has a sample of 2723 children
under the age of five which is just under half of the sample (6003) we have available in the LSMS 2006
survey (Larrea & Kawachi, 2005).
ii. The effect of income
The natural log of per capita household consumption is highly significant (always has a p<0.001) in every
type of model, be it OLS, multilevel or logistical multilevel, and when paired with the provincial, county
or parish level Gini coefficient. Income as measured by consumption at the household level has a strong
beneficial effect on the z-score of height-for-age and on the probability of being malnourished. The
concave effect that income has on health is captured in this variable as it is logged. The magnitude of the
effect is around 0.140 in the multilevel models, and around -0.049 in the marginal effects of the logistical
multilevel models.
This implies that a one unit change in Ln of consumption will increase the z-score by less than a one unit
change in the Gini coefficient. Additionally, a one unit change in the Ln of consumption will reduce the
risk of stunting by less than what a one unit increase in the Gini coefficient will reduce this risk by. It is
obvious that the units of influence this outcome, therefore, in order to properly compare these effects we
need to estimate standardized beta coefficients.
31
We also incorporate the mean consumption of the area of the model18
so as to capture the possible
externalities of having wealthier neighbors on access to healthcare or on individual behaviours. This
variable is significant and has a deleterious effect in the OLS models. However, the significance
disappears in the multilevel models and in the province and county logistical multilevel models. It would
seem that this variable was capturing the effect of unobserved heterogeneity in the OLS models.
Notwithstanding, It is significant in the parish logistical multilevel model and has a beneficial effect.
Therefore, there seems to be a positive externality, measured as a marginal effect of 0.001, of a higher
level of mean consumption of the parish on the child’s risk of being stunted.
Additionally, we control for the natural log of GDP of the specific geographic area19
in order to
incorporate the contextual effect of aggregate income and production. We find that the GDP has a
significant beneficial effect in every OLS model, and when measured at the province level in the
multilevel and logistical multilevel models. This suggests that there are potential positive externalities of
living in a province with a higher level of aggregate production, possibly because a higher level of GDP
may allow some provinces to better finance relevant social welfare and health services. Given the Ln of
household consumption and the Ln of GDP are measured in the same units we can see how the effect of
household consumption is larger and more significant than the effect of aggregate production, as a one
unit increase in monthly household consumption increases the z-score by around 0.14 and reduces the risk
of stunting by -0.049 while a one unit increase in annual provincial aggregate production increases the z-
score by 0.07 and reduces the risk of stunting by -0.025.
Finally, we measure the head count for poverty in the specific geographic area of each model. This
variable has a significant effect in the OLS models, however, it loses all significance in the multilevel
models. Again, perhaps this variable was capturing the effect of some unobserved heterogeneity in the
OLS models. However, it has a detrimental effect which is significant at the 10% level (p<0.1) when
measured at the parish level in the logistical multilevel models. Therefore there is partial evidence that the
level of poverty in a parish has a deleterious effect on the probability of stunting. Nevertheless, this effect
is very weak and in general it would seem that the incidence of poverty does not have a significant effect
on the nutrition of the child or on the risk of being stunted.
iii. Other effects
We discuss the effects of other regressors by grouping them in the same four categories as above:
characteristics of child, the mother, the household and contextual variables.
a. Individual characteristic of the child
18 Be it province, county or parish for the mean consumption, and be it province or county for GDP. 19 Although we do not have data for the GDP at the parish level so in this model we include the county GDP.
32
Most of the variables describing the characteristics of the child are significant. The gender is both highly
significant (p<0.001) in the OLS, multilevel and logistic multilevel regression. This implies that girls
have z-score on average 0.16 points higher and have a -0.051 lower risk of being stunted than boys. The
cubed effect of the age of the child seems to imply that younger children have a higher z-score on average
and that they are at a lower risk of being stunted. A one unit increase in age (measured in months) will
reduce the z-score by -0.149 initially, although this effect will fade out and inverse, then that inversed
effect will also fade out and inverse. A one unit increase in age will increase the risk of being stunted by
0.056 although this effect will also fade in a similar way. Children who are younger tend to still be
breastfeeding which may help maintain their growth patterns within the normal range, however, as they
get older they stop breastfeeding and perhaps at this moment they become more vulnerable to fall outside
of the normal growth range. This is probably the reason why we find that the number of months of
breastfeeding is not a significant variable; we suspect this effect is being absorbed by the age variable.
Children who are born with low birthweight have z-scores, on average, 0.603 points lower and have a
0.195 higher risk of being stunted than children born with normal birthweight. We find this variable is
significant (p<0.001) and deleterious in the OLS, multilevel and logistic multilevel models. This implies
that the pathway between psychosocial stress, prenatal maternal stress and LBW may be in action
although we are not testing it directly here. This evidence seems to indicate that there may be long-term
effects of LBW. Finally, the proportion of vaccines is not significant in any model except for the
multilevel provincial model where the effect is significant (p<0.05) and beneficial.
b. Characteristics of the mother
In the group of variables which describe the characteristics of the mother we can see that the years of
schooling has a significant (p<0.001) and beneficial effect in the OLS, multilevel and logistical multilevel
models. A unit increase in the years of schooling will increase the z-score of the child by 0.029 and
reduce the probability of stunting by -0.008. Additionally, the age of the mother has a similar result,
where it is significant (p<0.001) and beneficial to both dependent variables in each type of model. A one
unit increase in the age of the mother will increase the z-score by 0.01 and reduce the risk of stunting by -
0.003. This indicates that the nutrition of the child improves and the risk of stunting is reduced when the
mothers are older and more educated. We also find a significant and deleterious relation between the z-
score/risk of stunting and the fertility of the mother in the OLS, multilevel and logistical multilevel
models. It seems that the a one unit increase in fertility reduces the z-score by -0.699 points and increases
the risk of stunting by 0.338. The more children a mother has the worse the nutrition and the risk of
stunting of those children.
33
The effect of the access the mother has to prenatal, natal and postnatal healthcare is captured only by the
dummy for having a physician or obstetrician present at birth and whether the mother had a caesarian
section. Both have a beneficial effect on the nutrition of the child, and are the only variables of a list of
access to healthcare variables which were significant. However, the former is only significant in the
logistical models with a reduction in the risk of stunting of -0.042, and the latter may be capturing access
to certain healthcare services which are not widely accessible rather than the effect of the caesarean
section per se. This variable increases the z-score by 0.105 and reduces the risk of stunting by -0.03.
Finally, the dummy variable for the mother being underemployed is not significant in the OLS models,
however, it is significant and deleterious in the multilevel models and in the county and parish logistic
multilevel models. This demonstrates that the precarious working conditions of the mother affect the z-
score by -0.089 and increases the risk of stunting by 0.02.
c. Characteristics of the household
At the household level none of the included variables are significant in the OLS models except for the
square root of the number of children under the age of 12 in the household and, in the provincial model,
the access to government cash transfers. In the multilevel models, the variable for cash transfers is
rendered insignificant and the variable for the provision of nutritional supplements on behalf of the
government is rendered significant and deleterious. We assume that these nutritional supplements are
distributed to the households which need it the most, therefore, this sign captures the selection bias of this
program and not the effect of the treatment. In these multilevel models we also find that the housing
conditions index is significant and deleterious, however, it loses all significance in the logistical
multilevel models. Therefore, poor living conditions may reduce the z-score of the child by -0.07 points
however it does not increase the probability of stunting. We also find that the square root of the number of
children under 12 is significant and deleterious in the multilevel models and in the logistical multilevel
models. A one unit increase in this variable will reduce the z-score by -0.18 points and increase the risk of
stunting by 0.051.
Finally, we find a significant beneficial effect of the years of work experience of the head of the
household in the county and parish multilevel models and all the logistical multilevel models. A one unit
increase in this variable will increase the z-score by 0.003 and reduce the risk of stunting by -0.001. This
demonstrates that there is a beneficial effect of work stability after controlling for income and household
conditions. Therefore, this variable seems to be capturing an effect of work stability operating beyond
these parameters. We suspect that it may allow for a less stressful household environment which may be
conducive to lower levels of stress and violence against women and children.
34
d. Other contextual variables
In the OLS models the additional contextual variables are all significant except for the number of doctors
per every 10000 individuals. This variable is neither significant in the multilevel models nor in the
logistical multilevel models. The rate of higher education attendance has a significant and beneficial
effect in the multilevel and in the parish logistical multilevel models. A one unit increase in this variable
leads to an increase of around 0.3 points in the z-score and a reduction of -0.472 in the risk of being
stunted. This probably arises from the positive externalities on child care perceptions and behaviours of
living in a more educated community.
The Gini of land has a significant detrimental effect only in the OLS models; it loses its significance in
the multilevel models, and in the logistical multilevel models. The significance of the Gini of land in the
OLS models might have been capturing some of the unobserved heterogeneity which is controlled for in
the multilevel models.
The food consumption index, which gives high scores to parishes where there is a high consumption of
carbohydrates and a low consumption of proteins and micronutrients from fruits and vegetables, is
significant and deleterious multilevel and logistical multilevel model. A one unit increase in this index
may be interpreted as a parish with higher barriers to proper nourishment and food security. The effect of
this increase is a -0.071 reduction in the z-score and a 0.02 increase in the risk of stunting.
Finally we have a dummy variable for rural amazon and a dummy variable for rural areas. In the OLS
models, the dummy of rural amazon has a significant beneficial effect. The same is true in the multilevel
models and in the provincial and county logistical multilevel models. On average, the z-score is 0.44
points higher and the risk of stunting is around -0.1 lower than in the rest of the country. The dummy for
rural areas has a significant detrimental effect in the OLS and multilevel models where children have a z-
score between -0.12 and -0.232 points lower than the rest of the county, depending on the model. In the
logistical multilevel models this variable is no longer significant. The rural amazon areas are, as we
mentioned above, mainly the home of relatively isolated indigenous communities as well as the place
where there are oil extraction activities (Finer, et al., 2009). These indigenous communities may have
better nourished children because they have access to proper nutrition within this vastly biodiverse
environment or perhaps they have had better access to health care as a product of a increased contact with
western society through the introduction of oil companies. In the rest of the rural areas, however, children
have worse nutrition probably because of the large degree of isolation that still exists when living in rural
areas in Ecuador.
e. Fixed effects
35
The fixed effects we include are ethnicity variables which have three categories, mestizo20
, indigenous21
and afro-Ecuadorian. Being indigenous, after controlling for all the individual and contextual variables
has a deleterious effect on the z-score in the multilevel models, and on the probability of being stunted the
logistical multilevel models. This is expected given the high prevalence of malnutrition among
indigenous children (51%) which is double that of the national incidence (26%) (LSMS, 2006). On
average, indigenous children have a z-score which is -0.159 lower than other children and have a 0.039
higher risk of being stunted. The effect of being afro-Ecuadorian is the opposite and this variable is
significant in every model. Afro-ecuadorian children have z-scores 0.27 points higher, on average, than
the rest of children and a -0.115 lower risk of being stunted. It may be interesting to investigate the social
interactions and contextual conditions that lead to these opposing effects further.
iv. Limitations
We do not control for the anthropometric measures of the mother or father because the data is not
available in the survey. This would allow us to capture the genetic traits of the children which may have
an effect on their z-score. Typically, we would like to see if there is a relation between the z-score of
height for age of the child and the body mass index of the mother and father. It might be that parents who
are underweight have children who are stunted, or parents who are overweight have children who are
stunted or both. This is the greatest limitation of this study given the importance of genetics in children’s
growth patterns.
Secondly, the lack of panel data on health indicators in Ecuador limits our capacity to perform a study
where the unobserved heterogeneity of individuals which does not change over time may be controlled
for with a fixed effects model. This would allow for a model which is closer to obtaining causal effects
rather than correlation coefficients. This type of data is unfortunately unavailable in Ecuador.
6. Conclusion
Chronic malnutrition is an important health issue because it affects 1 in 4 children in the world (De Onis,
et al., 2012), and is the root cause of just under half (45%) of deaths in children under the age of five
(Horton & Lo, 2013). Its effects are potentially long-term and create deficits in cognition and educational
achievements (Granthan-MacGregor, et al., 2007; Grantham-McGregor, et al., 2000; Walker, et al., 2000;
Walker, et al., 2007) playing an important role in the intergenerational transmission of poverty. In this
20 Defined as a mix between Caucasian and Indigenous. 21 Indigenous people are usually fluent in one of various indigenous languages as well as Spanish, whereas it is rare for non-
indigenous peoples to be fluent in such languages. Therefore it may be suggested that the fluency in these indigenous languages be used as a proxy for ethnicity. Specifically, we find an individual to be indigenous if she or he either states it directly, or if she or he states that the spoken language in the household is an indigenous one. We do this in order to avoid an underestimation of the indigenous population.
36
paper our objective is to measure the effect of inequality on stunting in children under the age of 5. We
regress the z-score of height-for-age22
using multilevel regressions; and the dummy for stunting23
using
logistic multilevel regressions against the provincial, county and parish Gini coefficients while controlling
for individual, maternal, household and contextual characteristics. Our results show that, the Gini
coefficient has a significant deleterious correlation on both the z-score of height-for-age and the
probability of being malnourished in every model except for the logistical parish (local) level model.
In order to explain our findings we propose that inequality will affect individual nutrition of children
through two pathways: psychosocial stress affecting mothers and social cohesion protecting their
empowerment. Specifically, we argue in our first pathway that inequality through psychosocial
mechanisms can foster chronic amounts of stress among pregnant women, increasing their levels of
CRH,24
which regulates fetal maturation and increases the risk of low birth weight (LBW) (Beydoun &
Saftlas, 2008; Camacho, 2008; Mansour and Rees, 2011). This affects children’s nutrition because LBW
is an important determinant of chronic child malnutrition (Marins & Almeida, 2002; Willey et. al., 2009;
Aerts, et al., 2004; El Taguri, et al., 2009; Adair & David, 1997). In our second pathway, we argue that
inequality may erode social cohesion (Wilkinson, 2000; Kawachi, et al., 1999; Wilkinson, 1996) which
may be associated with an erosion of women’s interpersonal empowerment (Speer, et al., 2001; Peterson
& Hughey, 2004). This erosion at the community level may be conducive to an erosion of the mothers
agency over household resources (Lupri, 1990; Antai, et al., 2014) which may refocus these resources
away from the health and nutritional needs of children (Charles & Kerr, 1988; DeVault, 1991; Maitra,
2004; Shroff, et al., 2009; Imai, et al., 2014; Ackerson & Subramanian, 2008).
We have attempted to measure the first pathway indirectly using the data that is available to us in the
survey. We find that LBW has a significant and deleterious correlation with the z-score of height for age.
This provides some partial evidence of the pathway although we cannot measure the effect inequality has
on stress or the way in which, through stress, it affects birth outcomes. We have also attempted to control
for household level stressors through the employment stability of the head of the household. Variables
which measure stressors at the household level directly are not available in the data. We found that the
years of employment of the head of the household has a significant and beneficial correlation. We suspect
that, given we are controlling for income and household conditions, this employment variable is
indicating a level of household financial stability which reduces the risk of stress and violence. However,
this is neither a perfect proxy nor a specific indication that the pathway is actually in effect.
22 The normalized 𝑧-score establishes the growth standard of children by defining a normal growth curve, see page
16 for formal definition. 23 Children with a z-score under -2, see page 16 for formal definition. 24 Croticotrophin-Releasing Hormone
37
This study contributes to the larger literature regarding the health-inequality relation on three levels.
Fisrtly, there is only a small percentaje of studies25
found in Lynch at al. (2004) which focus on child
health (23.7%). Most of these studies measure Infant Mortality, which unlike nutrition, does not play a
role in the intergenerational transmition of poverty. Secondly, only 10.2% of the studies26
include Latin
American countries which we consider a shortfall given it is one of the most unequal regions of the world
(Inter-American Development Bank, 2000) and perhaps the ideal testing ground for the effect of
inequality. Finally, only 9.2% of the studies27
measure the Gini coefficients at different levels. Given the
effects of the Gini may vary depending on the how we measure the small area of interest, measuring the
Gini coefficient at various levels of aggregations allows a more profound analysis of inequality on health.
In the Wilkinson and Pickett review (2006) there are 45 international, 58 state levels, 25 county level and
40 small area level studies. As we can see there are relatively less studies performed at the county level
while mostly studies focus on the state level. Additionally, the distribution of the supportive, mixed and
unsupportive evidence favours the supportive at the state level (S: 51.7%, M: 25.8%, U: 22.4%) while it
is fairly balanced at the small area level (S: 30%, M: 35%, U: 35%). In this study, we find that the
magnitud of the Gini coefficient is smaller as the areas over which it is measured decreases in size.
Therefore, in order to fully assess the impact of inequality it is important to take different levels of
aggregation as we have done.
This study has the important limitations: anthropometric data on the nutritional status of parents is
lacking, and we do not have panel data. The nutritional status of the parents would allow us to control for
the genetic factors which may influence the growth patterns of children. It might be that parents who are
underweight have children who are stunted, or parents who are overweight have children who are stunted
or both. Secondly, the lack of panel data on health indicators in Ecuador limits our capacity to control for
unobserved heterogeneity which stays constant over time with a fixed effect model. The only data
available over time would be aggregated data from the census which would amount to an ecological
study. However, this type of study may be subject to the ecological fallacy28
and does not allow us to
disentangle the effect of inequality from the effect of aggregate income (Piantadosi, et al., 1988).
Some policy implications from our main findings are outlined here. Firstly, a redistributive policy which
will allow a reduction of poverty and inequality will help improve the nutritional health of children,
particularly if inequality at the county and provincial level is reduced. Secondly, the education of women
in general and of indigenous women in particular, is fundamental to the improvement of child nutrition.
25 Studies where the information was specified. 26 Studies where the information was specified. 27 Studies where the information was specified. 28 The incorrect equating of between group correlation with within group correlations.
38
Access to higher education in the local areas is also important in that it probably has positive externalities
for the local community. Thirdly, the access to obstetricians or doctors at the movement of birth is
significant and beneficial, giving force to the argument that targeted pre-natal, natal and post-natal health
services are best. Additionally, avoiding low birth weight is very important although the number of
medical visits during the pregnancy was not a significant variable. In this sense, there might be other
factors such as stress or violence which are affecting the health of the mother during the pregnancy.
Policy should focus on improving awareness of the effects of these conditions on the health of children.
Given the fertility of the mother is significant and detrimental and the age of the mother is significant and
beneficial, we also think that the accessibility of birth control is also fundamental given it has the
potential to reduce the number of children a woman has and augment the age at which a women has her
first child. Finally, access to a diversified diet had a large impact. Policy should focus on increasing food
security and reducing food poverty in certain areas in order to increase the amount of micronutrients and
protein available for consumption. Apparently, this type of policy would be more effect than direct cash
transfers which does not have a significant effect after controlling for income. We suspect that most social
programs will improve in effectiveness as the level of inequality diminishes.
39
Bibliografía
Ackerson, L. K. & Subramanian, S., 2008. Domestic violence and chronic malnutrition among women
and children in India. American Journal of Epidemiology, 167(10), pp. 1188-1196.
Adair, L. S. & David, K. G., 1997. Age-Specific Determinants of Stunting in Filipino Children. The
Journal of Nutrition, Volumen 127, pp. 314-320.
Adams, A. E., Sullivan, C. M., Bybee, D. & Greeson, M. R., 2008. Development of the scale of economic
abuse. Violence Against Women, 14(5), pp. 563-588.
Aerts, D., Drachler, M. d. L. & Justo Giugliani, E. R., 2004. Determinants of Growth Retardation in
Southern Brazil. Cadernos de Saúde Pública, 20(2), pp. 1182-1190.
Agren, G., 2003. Sweden´s new public health policy. Stockholm: Swedish National Institute of Public
Health.
Ahern, J., Pickett, K., Selvin, S. & Abrams, B., 2003. Preterm birth among African American and white
women: a multilevel analysis of socioeconomic characteristics and cigarette smoking. Journal of
Epidemiology & Community Health, Volumen 57, pp. 606-611.
Almond, D., Chay, K. Y. & Lee, D. S., 2005. The Costs of Low Birth Weight. The Quarterly Hournal of
Economics, 120(3), pp. 1031-1082.
Antai, D., Antai, J. & Anthony, D. S., 2014. The relationship between socio-economic inequalities,
intimate partner violence and economic abuse: A national study of women in the Philippines. Global
Public Health, 9(7), pp. 808-826.
Basu, K., 2006. Gender and say: A model of household behaviour with endegenously determined balance
of power. The Economic Journal, Volumen 116, pp. 558-580.
Behrman, J. R. & Rosenzweig, M. R., 2004. Returns to Birthweight. The Review of Economics and
Statistics, 86(2), pp. 586-601.
Bennett, J., 2003. Investment in population health in five OECD countries, Paris: OECD Health Working
Paper no. 2.
Berkman, L. & Glass, T., 2000. Social integration, social networks, social support and health. En: L.
Berkman & I. Kawachi, edits. Social Epidemiology. New York: Oxford University Press, pp. 137-177.
Beydoun, H. & Saftlas, A., 2008. Physical and mental health outcomes of prenatal maternal stress in
human and animal studies. Pediatric and Perinatal Epidemiology, 22(5), pp. 438-466.
Black, S., Devereux, P. & Salvanes, K., 2007. From the cradle to the labor market? The effect of birth
weight on adult outcomes. The Quarterly Journal of Economics, 122(1), pp. 409-439.
Blakely, T. A., Lochner, K. & Kawachi, I., 2002. Metropolitan area income inequality and self-related
health. Social Science & Medicine , pp. Vol. 51, 1343-1350.
40
Browning, C., 2002. The span of collective efficacy: extending social disorganization theory to partner
violence. Journal of Marriage and Family, Volumen 64, pp. 833-850.
Caetano, R., Milder, S. & Harris, T., 2010. Neighborhood characteristics as predictors of male to female
and female to male partner violence. Journal of Interpersonal Violence, Volumen 25, pp. 1986-2009.
Camacho, A., 2008. Stress and birth weight: evidence from terrorist attacks. American Economic Review,
98(2), pp. 511-515.
CEPAL Division de Desarrollo Economico, Division de Estadísticas, 2014. Social Panorama of Latin
America 2013. Santiago de Chile: ECLAC.
CEPAL División de Desarrollo Social y la Division de Estadísticas, 2013. Panorama Social de América
Latina, s.l.: CEPAL.
CEPAL, 2011. CEPALSTAT. [En línea]
Available at: http://websie.eclac.cl/sisgen/ConsultaIntegrada.asp
[Último acceso: 06 06 2013].
Charles, N. & Kerr, M., 1988. Women, Food and Families. Manchester: Manchester University Press.
Couzin, J., 2002. Quirks of fetal environment felt decades later. Science, Volumen 296, pp. 2167-2169.
Currie, J. & Hyson, R., 1999. Is the IMpact of Health Shocks Cushioned by Socioeconomic Status? The
Case of Low Birthweight. The American Economic Review, 89(2), pp. 245-250.
Davey Smith, G. & Egger, M., 1996. Commentary: Understanding it all - Health, meta-theories, and
mortality trends. British Medical Journal, pp. Vol. 313, No. 7072, 1584-1585.
De Onis, M., Blossner, M. & Borghi, E., 2012. Prevalence and trends of dtunting among pre-school
children 1990-2020. Public Health Nutrition, Volumen 15, pp. 142-148.
Deaton, A., 2003. Health, inequality, and economic development. Journal of Economic Literature, pp.
Vol. XLI, 113 – 158.
DeVault, M., 1991. Feeding the Family: The Social Organization of Caring as Gendered Work. Chicago:
University of Chicago Press.
Diez-Roux, A., 2000. Multilevel Analysis in Public Health Research. Annual Review of Public Health,
Volumen 21, pp. 171-192.
Diez-Roux, A. V., 1998. Bringing context back into epidemiology: Variables and fallacies in multilevel
analysis. American Journal of Public Health, 88(2), pp. 216-222.
El Taguri, A. y otros, 2009. Risk factors for stunting among under-fives in Libya. Public Health
Nutrition, 12(8), pp. 1141-1149.
Elbers, C., Lanjouw, J. O. & Lanjouw, P., 2003. Micro-Level Estimation of Poverty and Inequality.
Econometrica, pp. Vol. 71, No. 1, 355-364.
41
Elbers, C., Lanjouw, P. & Leite, P., 2008. Brazil within Brazil: Testing th epoverty map methodology in
Minas Gerais.
Ellison, G. T., 2002. Letting the Gini out of the bottle? Challenges facing the relative income hypothesis.
Social Science & Medicine, Volumen 54, pp. 561-576.
Eskenazi, B. y otros, 2007. Low birthweight in New York city and upstate New York following the
events of September 11th. Human Reproduction, 22(1), pp. 3013-3020.
Fafchamps, M., Kebede, B. & Quisumbing, A., 2009. Intra-household welfare in rural Ethiopia. Oxford
Bulletin of Economics and Statistics, 71(4), pp. 567-599.
Farrow, A., Larrea, C., Hyman, G. & Lema, G., 2005. Exploring the spatial variation of food poverty in
Ecuador. Food Policy, pp. Vol. 30, 510-531.
Finer, M. y otros, 2009. Ecuador's Yasuní Biophere Reserce: a brief modern history and conservation
chanllenges. Environmental Research Letters, 4(3).
Freire, W. B. y otros, 2014. The double burden of undernutrition and excess body weight in Ecuador.
American Journal of Clinical Nutrition, 100(6), pp. 1636S-46S.
Grantham-McGregor, S. M., Walker, S. P. & Chang, S., 2000. Nutritional deficiencies and later
behavioural development. Proceedings of the Nutrition Society, pp. Vol. 59, 47-54.
Granthan-MacGregor, S., Cheung, Y. B. & Cueto, S., 2007. Child develioment in developing countries 1
- Developmental potential in the first 5 years for children in developing countries. Lancet, 369(9555), pp.
60-70.
Health Canada, 1999. Toward a healthy future: Second report on health of Canadians. Ottawa: Minister
of Public Works and Government Services Canada.
Holzman, C. y otros, 2001. Pregnancy outcomes and community health: the POUCH study of preterm
delivery. Paediatric and perinatal Epidemiology, 15(2), pp. 136-158.
Horton, R. & Lo, S., 2013. Nutrition: a quintessential sustainable development goal. The Lancet,
382(9890), pp. 371-372.
Howden-Chapman, P. & Tobias, M., 2000. Social inequalities in health: New Zealand 1999. Wellington:
Ministry of Health.
Hsieh, C. & Pugh, M., 1993. Poverty, income inequality, and violent crime: a meta-analysis of recent
aggregate sata studies. Criminal Justice Review, Issue 18, pp. 182-202.
Imai, K. S., Annim, S. K., Kulkarni, V. S. & Gaiha, R., 2014. Women's Empowerment and Prevalence of
Stunted and Underweight Children in Rural India. World Development, Volumen 62, pp. 88-105.
Inter-American Development Bank, 2000. Development beyond Economics. Economic and social
progress in Latin America, Washington: IADB.
42
Janghorbani, M., Stenhouse, E., Millward, A. & Jones, R. B., 2006. Neighborhood deprivation and
preterm birth in Plymouth, UK. The Journal of Maternal-Fetal & Neonatal Medicine, 19(2), pp. 85-91.
Kaplan, G. y otros, 1996. Inequality in income and mortality in the United States: analysis of mortality
and potential pathways. BMJ, Volumen 312, pp. 199-103.
Karlsson, M., Nilsson, T., Lyttkens, C. H. & Leeson, G., 2010. Income inequality and health: Importance
of a cross-country perspective. Social Science & Medicine, pp. Vol. 70, 875 - 885.
Kawachi, I. & Kennedy, B., 1999. Income inequality and health: Pathways and mechanisms. Health
Services Research, 34(1), pp. 215-227.
Kawachi, I., Kennedy, B., Lochner, K. & Prothrow-Stith, D., 1997. Social capital, income inequality and
mortality. American Journal of Public Health, 87(9), pp. 1491-1498.
Kawachi, I., Kennedy, B. & Wilkinson, R., 1999. The Society and Population Health Reader: Income
inequality and health Vol 1. New York: New Press.
Larrea, C., 2002. Social inequality and child malnutrition in eight Latin American countries. Unpublished
paper, Harvard Center for Society and Health.
Larrea, C. & Freire, W., 2002. Social inequality and child malnutrition in four Andean countries. Pan
American Journal of Public Health, pp. Vol. 11 (5/6), 356-364.
Larrea, C., Freire, W. & Lutter, C., 2001. Equidad desde el principio: Situación nutricional de los niños
ecuatorianos, Washington, DC: Organización Panamericana de la Salud.
Larrea, C. & Kawachi, I., 2005. Does economic inequality affect child malnutrition? The case of Ecuador.
Social Science and Medicine, pp. Vol. 60, 165-178.
Larrea, C. & Montenegro Torres, F., 2006. Ecuador. En: Ingigenous peoples, poverty and human
development in Latin America 1994-2004. New York: Palgrave MacMillan, pp. 67-105.
Lupri, E., 1990. Male violence in the home. En: C. Mckie & K. Thompson, edits. Candian Social Trents.
Ottawa: Minister of Supply and Services Canada and Thompson Education Publishing Inc., pp. 170-172.
Lynch, J., 2000. Income Inequality and Health: Expanding the Debate. Social Science and Medicine, pp.
Vol. 51, No. 7, 1001-1005.
Lynch, J. y otros, 2004. Is income inequality a determinant of population health? Part 1. A systematic
review. The Milbank Quarterly, pp. Vol. 82, No. 1, 5-99.
Lynch, J. y otros, 2001. Income inequality, the psychosocial environment, and health: Comparisons of
wealthy nations. Lancet, pp. Vol. 358, 194-200.
Lynch, J., Due, P., Muntaner, C. & Davey Smith, G., 2000. Social capital - Is it a good investment
strategy for public health?. Journal of Epidemiology and COmmunity Health, pp. Vol. 54, No. 6, 404-408.
43
Lynch, J., Kapaln, G., Davey Smith, G. & House, J., 2000. Income inequality and mortality: Importance
to health of individual income, psychosocial environment, or material conditions. British Medical
Journal, pp. Vol. 320, No. 7243, 1200-1204.
Macinko, J. A., Shi, L., Starfield, B. & Wulu, J. T., 2003. Income inequality and health: A critical review
of the literatrue. Medical Care Research and Review, 60(4), pp. 407-452.
Maitra, P., 2004. Parental bargaining, health inputs and child mortality in India. Journal of Health
Economics, 23(2), pp. 259-591.
Mansour, H. & Rees, D., 2011. The effect of prenatal stress on birth weight: Evidence from the al-Aqsa
Intifada. IZA Discussion Paper No 5535.
Marins, V. & Almeida, M., 2002. Undernutrition prevalence and social determinants in children aged 0-
59 months, Niterói, Brazil. Annals of Human Biology, 29(6), pp. 609-618.
Marmot, M., 2004. Status syndrome: how your social standing directly affects your health and life
expectancy. London: Bloomsbury.
Marmot, M. & Wilkinson, R., 2006. Social determinants of health. Oxford: Oxford University Press.
McElroy, M., 1990. The empirical content of Nash-bargained household behaviour. Journal of Human
Resources, 25(4), pp. 559-583.
McElroy, M. & Horney, M., 1981. Nash-bargained household decision: Towards a generalisation of the
theory of demand. International Economic Review, 22(2), pp. 333-349.
Miller, D. L. & Paxson, C., 2006. Relative income, race and mortality. Journal of Health Economics,
Volumen 25, pp. 979-1003.
Nguyen, V.-K. & Peschard, K., 2003. Anthropology, inequality and disease: A review. Annual Review of
Anthropology, pp. Vol. 32, 447-474.
Olufunmilayo, F., 2008. Economic violence to women and girls - Is it receiving the necessary attention?.
Trauma Violence and Abuse, 9(3), pp. 167-177.
Organization of Economic Cooperation and Development, 2001. The well-being of nations: The role of
human and social capital. Paris: OECD, Centre for Educational Redearch and Innovation.
Park, C., 2007. Marriage market, parents' bargaining powers and child nutrition and education. Oxford
Bulletin of Economics and Statistics, 69(6), pp. 773-793.
Pearl, M., Braveman, P. & Abrams, B., 2001. The relationship of neighborhood socioeconomic
characteristics to birtheight among 5 ethnic groups in California. American Journal of Publich Health,
91(11), pp. 1808-1814.
Persson, G. y otros, 2001. Health in Sweden. The national public health report, 2001. Scandinavian
Journal of Public Health, pp. Vol. 29, Sup. 58, 199-218.
44
Peterson, N. A. & Hughey, J., 2004. Social cohesion and intrapersonal empowerment: gender as
moderator. Health Education Research, 19(5), pp. 533-542.
Piantadosi, S., Byar, D. P. & Green, S. B., 1988. The Ecological Fallacy. American Journal of
Epidemiology, 127(5), pp. 893-904.
Power, E. M., 2006. Economic abuse and intra-household inequities in food security. Canadian Journal
of Public Health, 97(3), pp. 258-260.
Preston, S., 1975. The changing relation between morality and level of economic development.
Population Studies, pp. Vol. 29, 231-248.
Programa Alimentate Ecuador, Ministerio de Inclusión Económica y Social, Ecuador, 2009. Lista de
alimentos de la Encuesta de Condiciones de Vida, Quito: Unpublished.
Putnam, R., Leonardi, R. & Nanetti, R., 1993. Making democracy work: civic traditions in modern Italy.
Princeton: Princeton University Press.
Raudenbush, S. & Bryk, A., 2002. Hierarchical linear models: Applications and data analysis methods.
Thousand Oaks: Sage Publications.
Rodgers, G., 1979. Income and inequality as determinants of mortality: An international cross-section
analysis. Population Studies, pp. Vol. 33, 343-351.
Satatistics Canada, Candian Centre for Justice Statistics, 2000. Family Violence in Canda: A Statistical
Profile 2000. Ottawa: Ministry of Industry.
Shroff, M. y otros, 2009. Maternal autonomy is inversely related to child stunting in Andhra Pradesh,
India. Maternal & Child Nutrition, 5(1), pp. 64-74.
Snijders, T. & Bosker, R., 1999. Multilevel analysis. London: Sage.
Speer, P., Jackson, C. & Peterson, N., 2001. The relationship between social cohesion and empowerment:
Support and new implications for theory. Health education & behavior, 28(6), pp. 716-732.
Steckel, R. H., 1995. Stature and the standard of living. Journal of Economic Literature, pp. Vol. 33, No.
4, 1903-1940.
Subramanian, S. V. & Kawachi, I., 2004. Income Inequality and Health: What Have We Learned So Far?.
Epidemiologic Reviews, pp. Vol. 26, 78–91.
Tarozzi, A. & Deaton, A., 2009. Using censu and survey data to estimate poverty and inequality for small
areas. The Review of Economic and Statistics, 91(4), pp. 733-792.
Turrell, G., Oldenburg, B., McHuffog, I. & Dent, R., 1999. Socioeconomic determinants of health:
Towards a national research program and a policy and interventional agenda. Canberra: Queensland
University of Technology, School of Public Health, Ausinfo.
45
Unidad de Análisis de Información del Ministerio de Coordinación de Desarrollo Social, 2012. Mapas de
Pobreza y Desigualdad del Ecuador, Quito: Secretaría Tecnica del Ministerio de Coordinación de
Desarrollo Social.
VanderEnde, K. E., Yount, K. M., Dynes, M. M. & Sibley, L. M., 2012. Community-level correlates of
intimate partner violence against women globally: A systematic review. Social Science & Medicine, Issue
75, pp. 1143-1155.
Wafstaff, A. & Van Doorslaer, E., 2000. Relative income, race and mortality. Annual Revision Public
Health, pp. Vol. 21, 543–567.
Walker, S. P., Grantham-McGregor, S. M., Powel, C. A. & Chang, S. M., 2000. Effects of growth
restriction in early childhood on growth, IQ, and cognition at age 11 to 12 years and the benefits of
nutritional supplementation and psychosocial stimulation. The Journal of Pediatrics, pp. Vol. 137, No. 1,
36-41.
Walker, S. P. y otros, 2007. Childe development: risk factors for adverse outcomes in developing
countries. The Lancet, 369(9556), pp. 145-157.
Weede, E., 1981. Income Inequality, Average Income and Domestic Violence. Journal of Conflict
Resolution, 25(4), pp. 639-654.
Wesserman, C. R. y otros, 1998. Socioeconomic status, neighborhood social conditions, and neural tube
defects. Americal Journal of Public Health, 88(11), pp. 1674-1680.
Wilkinson, R., 1992. Income distribution of life expectancy. British Medical Journal, pp. Vol. 304, No.
6820, 165-168.
Wilkinson, R., 1999. Health, hierarchy and social anxiety. En: N. Adler, M. Marmot, B. McEwen & J.
Stewart, edits. Socioeconomic status and health in industrial nations: social, psychological and biological
pathways. New York: New York Academy of Sciences, pp. 48-63.
Wilkinson, R. G., 1996. Unhealthy societies: the afflictions of inequality. London: Rourledge.
Wilkinson, R. G., 2000. Inequality and the social environment: a reply to Lynch et al. Journal of
Epidemiol Community Health, Issue 54, pp. 411-413.
Wilkinson, R. & Pickett, K., 2006. Income inequality and population health: a review and explanation of
the evidence. Social Science and Medicine, Volumen 62, pp. 1768-1784.
Wilkinson, R. & Pickett, R., 2009. Income inequality and social dysfunction. Annual Review of
Sociology, Volumen 35, pp. 493-511.
Willey, B. y otros, 2009. Socio-economic predictors of stunting in preschool children: a population-based
study from Johannesburg and Soweto. SAMJ, 99(6), pp. 450-456.
Williams, R., Feaganes, J. & Barefoot, J., 1995. Hostility and death rates in 10 US cities. Psychosom
Med, Issue 57, p. 94.
46
World Health Organization, 1997. WHO global database on child growth and malnutrition, Geneva:
WHO.
World Health Organization, 2013. WHO Child Growth Standards: Methods and development. [En línea]
Available at: http://www.who.int/childgrowth/publications/technical_report_velocity/en/index.html
World Health Organization, 2013. World Health Organization, Child Growth Standards. [En línea]
Available at: http://www.who.int/childgrowth/software/en/
Wright, E. & Benson, M., 2010. Immigration and intimate partner violence: exploring the immigrant
paradox. Social Problems, Volumen 57, pp. 480-503.
47
Appendix 1: Detailed explanation of our small area estimates methodology based on Elbers et al (2003)
Let W be an indicator of poverty or inequality based on the distribution of household-level consumption,
yh. Using the sample from the Ecuadorian LSMS (2006), a small but rich data sample, we can estimate
the joint distribution of yh and the covariates xh. This estimated distribution can be used to generate the
distribution of yh for any subpopulation of the 2010 census if the set of explanatory variables are
restricted to those which can also be found in this census, in turn allowing us to generate the conditional
distribution of W, its point estimate and its prediction error (Elbers, et al., 2003).
Consider equation (1), a linear approximation of the conditional distribution of ych, where c is a sample
cluster and h is household, where the vector of disturbances is u ∽ ℱ(0, Σ). To permit for a within cluster
correlation in disturbances we use the equation (2) specification where η and ε are independent of each
other and uncorrelated with xh (Elbers, et al., 2003).
(1) ln 𝑦𝑐ℎ = 𝐸[ln 𝑦𝑐ℎ|𝑥𝑐ℎ𝑇 ] + 𝑢𝑐ℎ = 𝑥𝑐ℎ
𝑇 𝛽 + 𝑢𝑐ℎ
(2) 𝑢𝑐ℎ = 𝜂𝑐 + 𝜀𝑐ℎ
Basically, an initial estimation of β in equation (1) is obtained using OLS and we denote the residuals of
this regression as uch. With consistent estimates of β the residuals ech can be used to estimate the
variance of εch (Elbers, et al., 2003).
(3) ��𝑐ℎ = ��𝑐. + (��𝑐ℎ − ��𝑐.) = ��𝑐 + 𝑒𝑐ℎ
Subsequently, we use this estimated distribution (1) to generate the expected value of W in a
subpopulation of the census which we denote v for village. Thus we write W(mv, Xv , β, uv) where mv is
the Mv-vector of household sizes in village v, Xv is the matrix of observable characteristics, and uv is the
vector of disturbances which is unknown and therefore estimated as we have explained. The expected
value of W is then μv = E[W|mv, Xv, ζv] where ζ is the vector of model parameters which includes the
disturbances. In constructing an estimator of μv we replace ζv with ζv. This gives us
μv = E[W|mv, Xv , ζv] which is often analytically intractable so simulation is used to obtain the estimator
μv (Elbers, et al., 2003).
The difference between μv and the actual level of W has three components and can be written as follows
(Elbers, et al., 2003)
(4) 𝑊 − �� = (𝑊 − 𝜇) + (𝜇 − ��) + (�� − ��)
The idiosyncratic error – (W − μ): the difference between the actual value and the expected value of W
arises from the unobserved component of consumption and increases as the size of the target population
shrinks which limits the degree of desegregation possible (Elbers, et al., 2003).
48
The model error – (μ − μ): given ζv are consistent estimators of ζv, μ is a consistent estimator of μ and
√s(μ − μ) d→ 𝒩(0, ΣM) as s ⟶ ∞. However, given that this component of the prediction error is
determined in equation (1), it does not change systematically with changes in the size of the target
population (Elbers, et al., 2003).
The computation error – (μ − μ): when simulation is used as a method of computation, this error has an
asymptotic distribution √R(μ − μ) d→ 𝒩(0, Σc) as R ⟶ ∞. Where R is the number of independent
random draws used for the simulation and therefore this error can be as small as the computational
resources allow (Elbers, et al., 2003).
In table 3 we present the median standard error, population and number of households in the parishes
within each of these sub-regions. As we can see, the number of households in every parish is well below
the 15000 mark established in Elbers, et al. (2003) which indicates that the total variance due to the
idiosyncratic component of the error might be somewhat above the ideal level (0.24). Nevertheless, the
median standard errors in this study tend to be smaller than those presented by Elbers et al. (2003) and the
population size is similar or larger than those found in Elbers et al. (2003).
Table 6 Population size and standard errors of Gini coefficient
Region
Median #households parish (1000s) Median Standard Error (parish) Median Population (parish) (1000s)
This study This study Elbers et al. (2003) This study Elbers et al. (2003)
Quito 4.2 0.0075 0.048 14.4 5.8 Guayaquil 5.3 0.0081 0.039 15.1 6.5 Urban Coast 3.6 0.0102 0.031 9.2 11 Rural Coast 1.1 0.0127 0.046 4.3 4.6 Urban Highlands 2.6 0.0066 0.026 8.2 10 Rural Highlands 0.8 0.0144 0.038 2.8 3.3 Urban Amazon 3.2 0.011 0.027 9.3 8 Rural Amazon 0.4 0.022 0.043 1.4 1.2 National Total 1 0.014
3.6
Source: Redrawn from Elbers, et al., 2003
In 2003, Elbers et al. use the LSMS of 1994, and take household per-capita expenditure as their indicator
of wellbeing. They construct populations of increasing sizes from a constant distribution 𝐺𝑣(𝑥, 𝑚) by
randomly drawing households from the census households in one particular region of the country, the
rural coast. For each population size the table shows the welfare estimations, the standard errors of the
predictions and the share of total variance due to the idiosyncratic component of the error. The
idiosyncratic error is important given that it is the component of the total error which increases with a
reduction in the target population. Elbers, et al. (2003), demonstrate, in table 3, that for a sample of 15000
households the idiosyncratic component, 𝑉𝑙, is small and there is little to gain from increasing the sample
size or moving to higher levels of aggregation (Elbers, et al., 2003).
Table 7 Simulation Results
Measure Estimated Values Number of Household
100 1000 15000 100000
49
Headcount µ 0.46 0.5 0.51 0.51 Total Standard Error 0.067 0.039 0.024 0.024 𝑉𝑙/Total Variance 0.75 0.24 0.04 0.02
General Entropy (0.5) µ 0.26 0.28 0.28 0.28 Total Standard Error 0.048 0.029 0.022 0.022 𝑉𝑙/ Total Variance 0.79 0.28 0.03 0.01
Source: Redrawn from Elbers, et al., 2003
As they have shown, when combining census and survey data, desaggregating to subregions and
estimating poverty or in our case the Gini coefficient and the per-capita household consumption, for
specific locations becomes possible. When they estimate their welfare indicators by parishes they
demonstrate, in table 4, that one can estimate the Gini coefficient using combined data for subpopulations
one hundredth the size of those one can estimate in survey data and obtain very similar prediction errors
(Elbers, et al., 2003).
50
Table 8 Improvement using combined data
Region
Sample Data Only (region) Combined Data (suregions)
(2) S.E. of Estimate
(3) Population
(1000s)
(4) S.E. of Estimate
Median
(5) Population
Median (1000s)
Rural Highlands 0.027 2509 0.038 3.3 Rural Cost 0.042 1985 0.046 4.6 Rural Amazon 0.054 298 0.043 1.2 Urban Highland 0.026 1139 0.026 10 Urban Cost 0.03 1895 0.031 11 Urban Amazon 0.05 55 0.027 8 Quito 0.033 1193 0.048 5.8 Guayaquil 0.027 1718 0.039 6.5
Source: Redrawn from Elbers, et al., 2003
51
Appendix 2: Equation (1) for Small Area Estimates
We construct a separate model for each subregion Dependent Variable: LNCONPCM Weighted by: FEXP Quito Guayaquil Urban Coast Rural Coast Urban Highlands Rural Highlands Urban Amazon Rural Amazon
_intercept_ 3.9244*** 4.69609*** 4.95897*** 4.24822*** 4.66441*** 3.81323*** 4.21719*** 5.12909***
(0.54644) (0.32232) (0.17108) (0.16086) (0.23849) (0.17533) (0.33065) (0.2845)
Access to higher education 0.90309***
(0.09756)
Average proportion of household with cement walls in statistical area
-0.14443***
0.13408***
(0.04016)
(0.03017)
Average proportion of houses with connection to public water disposal service
-0.16132** -0.12525**
(0.05754) (0.03788)
Average proportion of housing with exclusive toilets in statistical area -0.11827
-0.11533
0.22179**
(0.14344)
(0.06497)
(0.07307)
Average proportion of houses with garbage truck service in statistical area 1.231*
-0.12598*
(0.5303)
(0.05113)
Average proportion of houses with publicly provided drinking water in statistical area
0.05851
-0.32875**
(0.03787)
(0.11025)
Average proportion of persons per room in statistical area
-0.20641 -0.05895**
-0.06983*** 0.04303 -0.20006***
(0.15818) (0.02121)
(0.01585) (0.02425) (0.02644)
Average years of schooling in parish
0.04799**
(0.0176)
Dummy amplified nuclear family
-0.08971** 0.05536*
(0.02868) (0.02572)
Dummy Cuenca
0.16585***
(0.02294)
Dummy bamboo flooring or similar
-0.21931** -0.07458 -0.09438**
-0.06947**
(0.07354) (0.0439) (0.03403)
(0.02424)
Dummy for cement or brick flooring
-0.1966***
-0.08137***
(0.04597)
(0.02228)
Dummy for walls made of bamboo wood or similar
-0.17853** -0.06676*
(0.0547) (0.0309)
Dummy head of household affiliated to social security
0.08396** 0.0631** 0.11415**
0.1265***
(0.03172) (0.0237) (0.03546)
(0.03291)
Dummy head of household construction worker
-0.0573 0.32542*** -0.12684* 0.08557*** 0.12598*** 0.20716***
(0.04312) (0.06257) (0.04962) (0.02146) (0.02989) (0.05118)
Dummy head of household directive position 0.17944* 0.31372***
0.27233*
0.31744**
(0.07427) (0.067)
(0.12508)
(0.10396)
Dummy head of household employer 0.09348 0.15455*** 0.23413*** 0.19037*** 0.22634*** 0.16207*** 0.21176**
(0.05634) (0.04382) (0.02869) (0.02919) (0.03255) (0.03692) (0.06899)
Dummy head of household ethnic 0.07023 0.07685
-0.14128**
(0.05121) (0.04629)
(0.04509)
Dummy head of household female -0.08383*
-0.09081** 0.04279
(0.03465)
(0.02807) (0.02744)
Dummy head of household in hotel industry
0.12719* 0.12641
(0.04931) (0.07932)
Dummy head of household in manufacturing
0.05821
(0.03067)
Dummy head of household in retail sale
0.10744*** 0.04692 0.0947**
-0.0909 -0.39423***
(0.02441) (0.03929) (0.02895)
(0.06048) (0.10429)
Dummy head of household inactive
0.09164*
0.04816
0.4985***
(0.04158)
(0.03294)
(0.1199)
Dummy head of household marital status divorced/separated
-0.05765* 0.00655
(0.02748) (0.03115)
Dummy head of household marital status single
-0.13252*** -0.06892
-0.11749***
(0.03986) (0.04049)
(0.03471)
Dummy head of household non-qualified agricultural worker
0.0731*
-0.23459***
(0.0363)
(0.053)
Dummy head of household non-qualified worker
-0.04806*
(0.02252)
Dummy head of household other service position -0.15888 0.12451*
-0.11051
(0.08409) (0.05073)
(0.06869)
Dummy head of household over 65 years of age
-0.15334** -0.07733* -0.158***
(0.04989) (0.03123) (0.02902)
Dummy head of household public sector
0.38836 0.15076***
0.2044 -0.36949* 0.23886**
52
(0.22173) (0.03427)
(0.11238) (0.16818) (0.0731)
Dummy head of household salary worker -0.08119* -0.0754*
-0.08235** -0.08073**
-0.14238**
(0.0359) (0.03323)
(0.02572) (0.0247)
(0.05453)
Dummy head of household speaks native language
-0.03236
-0.10053*
(0.0292)
(0.04922)
Dummy head of household speaks native language and Spanish
0.0881
(0.0559)
Dummy head of household transportation
0.09764* 0.11166**
0.15424*** 0.12366** 0.17751*
(0.04331) (0.03431)
(0.03626) (0.04404)
Dummy head of household wholesale worker
0.10023* 0.09169* 0.24802*** 0.12789*** 0.19403***
(0.04689) (0.0401) (0.07271) (0.03801) (0.05837)
Dummy head of household widow/widower
-0.05469
-0.01949
(0.05432)
(0.03492)
Dummy head of household works in modern sector
0.05895**
0.09492*** 0.12742*
(0.02019)
(0.01993) (0.05373)
Dummy household garbage is burnt or buried
0.06808***
0.13293 0.11474*
(0.01873)
(0.08065) (0.04524)
Dummy household garbage is thrown in empty lot
0.29669*
(0.11947)
Dummy household that share or do not have toilet
0.04359
(0.02267)
Dummy household water connection outside the building and the property -0.27689*
-0.01256
-0.17965*** -0.38079**
(0.12968)
(0.01809)
(0.04664) (0.14088)
Dummy household with adobe walls
-0.15419*
(0.06985)
Dummy household water connection outside the building but inside the property
-0.07658* -0.0674* 0.1763**
(0.03296) (0.02751) (0.06746)
Dummy household with asbestos roof or similar -0.09016*
0.02289 0.00517 0.05291**
(0.03702)
(0.03795) (0.02097) (0.02006)
Dummy household with electric stove 0.47731**
0.97216*
(0.15683)
(0.42308)
Dummy household with palm/straw roof or similar
-0.08914*
-0.04628
(0.03572)
(0.08125)
Dummy household with room for rent
0.0358
(0.03725)
Dummy household with wood walls
-0.05554
(0.09368)
Dummy household wood/coal stove
-0.10355*** -0.23951** -0.17803*** -0.61124** -0.28142***
(0.02589) (0.08161) (0.02292) (0.19751) (0.05506)
Dummy housing with no electricity
-0.67314 0.0689
-0.38417* -0.10891**
(0.38913) (0.0757)
(0.17524) (0.03945)
Dummy housing with no telephone -0.22134*** -0.13318*** -0.23438*** -0.26983*** -0.19948*** -0.17905*** -0.19044*** -0.28064***
(0.03785) (0.02912) (0.02075) (0.04135) (0.02135) (0.02539) (0.04837) (0.08092)
Dummy housing provided in exchange for services
-0.07415** -0.07281** -0.09788*** -0.04755*
-0.12341*
(0.02395) (0.02215) (0.02516) (0.02301)
(0.05236)
Dummy housing with exclusive room for cooking
0.18247**
0.1422*
0.36858**
0.40627***
(0.06259)
(0.05986)
(0.11203)
(0.12162)
Dummy housing with latrine -0.95173*** -0.18963*
-0.04757
(0.27272) (0.08365)
(0.02945)
Dummy housing with no shower -0.20037*** -0.10793** -0.14549***
-0.01579 -0.11995*** -0.13628*
(0.04564) (0.03456) (0.02258)
(0.03132) (0.02232) (0.05606)
Dummy housing with other stove -0.14827
(0.17555)
Dummy housing with toilet and septic tank
0.04789 0.03785 0.12972***
0.0763***
(0.0315) (0.02089) (0.02104)
(0.02111)
Dummy incomplete nuclear family
0.04721
-0.01417 0.09195** 0.08409
(0.02525)
(0.02866) (0.02787) (0.05071)
Dummy indigenous head of household 0.08425
(0.09848)
Dummy metal zinc roof
-0.07327*
0.0561
0.0607
(0.0302)
(0.03515)
(0.04587)
Dummy tile flooring or similar
0.08308***
(0.02116)
Dummy precarious housing
0.41124**
-0.12429
(0.15054)
(0.07863)
Dummy rented housing -0.12113*** -0.08006** -0.07646**
-0.08938* -0.12164*
53
(0.03102) (0.0302) (0.02409)
(0.04269) (0.04872)
Dummy semi-precarious housing
0.13243** 0.08558* 0.04575
-0.12763***
(0.04624) (0.0334) (0.03728)
(0.03208)
Dummy tile flooring or similar
0.05854 0.19685*** 0.23252***
0.11607* 0.14214**
(0.05148) (0.02308) (0.04915)
(0.04679) (0.05326)
Elementary school attendance net rate
0.23752 -0.54376**
(0.16138) (0.18509)
Head of household education * dummy head of household formal sector 0.0108**
0.01024***
(0.00329)
(0.00183)
Head of household education * dummy head of household public sector
0.00409 0.00498*
-0.02533*
(0.00289) (0.00227)
(0.01004)
Head of household education * dummy head of household house worker
-0.02849**
-0.07134*
(0.00974)
(0.03342)
Head of household education * dummy head of household public sector
-0.02311
0.01864
(0.0149)
(0.01284)
Head of household education * head of household experience 0.00017 0.00078*** 0.00027** 0.00065*** 0.00046***
(0.00024) (0.00022) (0.0001) (0.00012) (0.000074955)
Head of household experience 0.00323
-0.00072 0.01974*
0.0009
(0.00436)
(0.00214) (0.00993)
(0.00104)
Head of household experience2 0 -0.00036*
-0.00093*
(0) (0.00014)
(0.00036)
Head of household experience3
0* 0 0.000011198**
(0)
(0)
Head of household schooling 0.01307 -0.03468*
0.03081*** 0.00616
0.02007***
(0.01875) (0.01688)
(0.00305) (0.00684)
(0.00584)
Head of household schooling2 0.00066 0.0012 0.00081*** 0.0004
0.00168*** 0.00078*
(0.00068) (0.00064) (0.0002) (0.00027)
(0.00043) (0.00033)
High school attendance net rate in parish 0.15497* -0.05731 0.05553 0.09346**
-0.07176*
(0.06223) (0.0506) (0.04261) (0.03176)
(0.03481)
Household water obtained from stream or similar
-0.13784***
0.09818** 0.17614
(0.03786)
(0.03553) (0.10576)
Household water obtained well
-0.13309*** 0.03924*
0.18506**
(0.03445) (0.01996)
(0.064)
Household with room for family business
0.03804
0.2489***
(0.03129)
(0.07349)
Household with toilet without septic tank, just dung up well
-0.02735
-0.1627*
0.09123
(0.07835)
(0.05648)
Ln(Income per-capita) 0.04942*** 0.17483***
0.14104***
0.14508*** 0.11786*** 0.17597***
(0.01444) (0.02635)
(0.01217)
(0.01589) (0.0289) (0.03269)
Percentages of houses in parish with parquet floors or similar
-0.39243
0.08402 -0.26581
(0.24413)
(0.07599) (0.20628)
Rate of literacy in statistical area
0.37477*
0.40517 0.21459**
(0.14607)
(0.2192) (0.07405)
Rooms per person 0.22059*** 0.25524*** 0.21138*** 0.16251*** 0.21087*** 0.18977*** 0.14312** 0.09979**
(0.02652) (0.03094) (0.01879) (0.01775) (0.01513) (0.01748) (0.04331) (0.03445)
Square root of number of basic needs met -0.08573*** -0.06066 -0.05829** -0.13819*** -0.11805*** -0.08635** -0.21112*** -0.1249***
(0.02511) (0.03152) (0.01799) (0.0273) (0.02079) (0.02689) (0.04291) (0.03662)
Square root of number of hours of work of head of household
-0.02393** 0.02043***
-0.00328
0.03463** 0.02253
(0.00832) (0.0048)
(0.00457)
(0.013) (0.01175)
Square root of number of people in household -0.38042*** -0.28734*** -0.33855*** -0.33752*** -0.2499*** -0.30488*** -0.34108*** -0.43351***
(0.05296) (0.04401) (0.03044) (0.03138) (0.03465) (0.03308) (0.0697) (0.05541)
Square root of number of people under 12 in household -0.08027** -0.03039 -0.11979*** -0.11686*** -0.1102*** -0.04395* -0.14246** -0.10295*
(0.03088) (0.02715) (0.01808) (0.02152) (0.01927) (0.02126) (0.04559) (0.04293)
University attendance net rate in parish -0.05373 0.20962** -0.0629
0.18251*** 0.17734** 0.47892**
(0.07623) (0.06353) (0.03785)
(0.03706) (0.05998) (0.18259)
Water provision by water truck
0.17168***
0.20028***
(0.04711)
(0.05823)
R2 0.77657 0.75135 0.69353 0.62719 0.69593 0.63059 0.8007 0.80229 N 878 1010 2566 2154 2314 3008 388 592
54
Appendix 3: Point estimation and standard errors of Gini coefficients estimations for provinces Region Provincial code Gini coefficient Standard error Population Number of HH
Quito (county) 1701 0.422 0.005 1933579 566115
Guayaquil (county) 901 0.401 0.007 1584401 589778
Urban Coast (excluding Guayaquil) 2 0.393 0.009 8766 3465
Urban Coast (excluding Guayaquil) 3 0.391 0.010 25560 8795
Urban Coast (excluding Guayaquil) 5 0.381 0.011 18015 6524
Urban Coast (excluding Guayaquil) 6 0.367 0.009 6220 2308
Urban Coast (excluding Guayaquil) 7 0.389 0.007 327899 119633
Urban Coast (excluding Guayaquil) 8 0.428 0.008 181985 69939
Urban Coast (excluding Guayaquil) 9 0.416 0.008 560022 214528
Urban Coast (excluding Guayaquil) 11 0.385 0.012 8329 3113
Urban Coast (excluding Guayaquil) 12 0.394 0.007 309347 114847
Urban Coast (excluding Guayaquil) 13 0.403 0.007 526870 193795
Urban Coast (excluding Guayaquil) 17 0.368 0.011 4552 1465
Urban Coast (excluding Guayaquil) 20 0.370 0.009 14601 5447
Urban Coast (excluding Guayaquil) 23 0.401 0.007 200708 69863
Urban Coast (excluding Guayaquil) 24 0.403 0.010 144331 49525
Rural Coast 2 0.353 0.009 31831 8387
Rural Coast 3 0.358 0.009 19543 4969
Rural Coast 4 0.370 0.015 6056 1351
Rural Coast 5 0.335 0.008 32110 8105
Rural Coast 6 0.339 0.017 4212 1118
Rural Coast 7 0.332 0.006 141682 39381
Rural Coast 8 0.342 0.006 240296 58969
Rural Coast 9 0.314 0.006 494855 136403
Rural Coast 10 0.345 0.012 8207 1928
Rural Coast 11 0.369 0.009 55279 14722
Rural Coast 12 0.313 0.005 318389 85089
Rural Coast 13 0.330 0.006 562389 144174
Rural Coast 17 0.344 0.007 47164 11460
Rural Coast 20 0.361 0.017 5720 1714
Rural Coast 23 0.337 0.007 85842 21646
Rural Coast 24 0.328 0.007 99719 24786
Rural Coast 90 0.319 0.006 31066 7834
Urban Highlands (excluding Quito) 1 0.371 0.004 273644 95965
Urban Highlands (excluding Quito) 2 0.375 0.005 21967 8391
Urban Highlands (excluding Quito) 3 0.370 0.004 38131 12784
Urban Highlands (excluding Quito) 4 0.362 0.004 58635 19304
Urban Highlands (excluding Quito) 5 0.366 0.004 74060 24665
Urban Highlands (excluding Quito) 6 0.361 0.004 116097 41975
Urban Highlands (excluding Quito) 10 0.380 0.004 187589 62345
Urban Highlands (excluding Quito) 11 0.377 0.004 157668 53480
Urban Highlands (excluding Quito) 17 0.381 0.004 133364 42489
Urban Highlands (excluding Quito) 18 0.356 0.003 163239 55994
Rural Highlands 1 0.400 0.007 306371 87950
Rural Highlands 2 0.444 0.010 96118 26867
Rural Highlands 3 0.401 0.007 100620 30827
Rural Highlands 4 0.386 0.009 76723 22245
Rural Highlands 5 0.426 0.009 235744 62505
Rural Highlands 6 0.421 0.008 256464 77644
Rural Highlands 10 0.420 0.008 136546 36813
Rural Highlands 11 0.403 0.009 141291 42390
Rural Highlands 17 0.435 0.008 358677 99396
Rural Highlands 18 0.385 0.006 278716 81438
Rural Highlands 23 0.382 0.009 8945 2514
Urban Amazon 14 0.397 0.009 26169 8657
Urban Amazon 15 0.397 0.007 22670 7428
Urban Amazon 16 0.391 0.008 30185 10249
Urban Amazon 19 0.378 0.007 16458 5275
Urban Amazon 21 0.379 0.007 49623 17991
Urban Amazon 22 0.394 0.009 38886 13992
Rural Amazon 14 0.560 0.010 81415 24128
Rural Amazon 15 0.524 0.010 59677 14910
Rural Amazon 16 0.540 0.010 33408 9212
Rural Amazon 19 0.491 0.012 53595 15710
Rural Amazon 21 0.485 0.013 83056 24791
Rural Amazon 22 0.511 0.012 64842 17385
55
Appendix 4: Housing conditions index
We use the first component as our housing conditions index. The index increases as the living conditions improve.
Component Matrixa
Component
1 2
Dummy houses with a sewage connection .784 -.228 Dummy houses with public garbage collection services .765 -.081 Dummy houses with exclusive washroom .726 -.211 Dummy houses with electricity .465 .536 Dummy houses with viable walls .622 .368 Dummy houses with viable floors .694 .207 Dummy houses with a water connection .770 -.097 Dummy houses with viable roof .232 .605 Dummy houses with phone connection .657 -.287 Dummy houses with overcrowding -.302 .349
Extraction Method: Principal Component Analysis. a. 2 components extracted.
Total Variance Explained
Component
Initial Eigenvalues Extraction Sums of Squared Loadings
Total % of Variance Cumulative % Total % of Variance Cumulative %
1 3.980 39.803 39.803 3.980 39.803 39.803 2 1.149 11.489 51.293 1.149 11.489 51.293 3 .935 9.352 60.645 4 .875 8.748 69.393 5 .766 7.656 77.049 6 .673 6.726 83.775 7 .482 4.821 88.596 8 .448 4.484 93.080 9 .371 3.713 96.792 10 .321 3.208 100.000
Extraction Method: Principal Component Analysis.
Results of principal component analysis of housing conditions
Sub-region Mean N Std. Deviation
Quito 0.9154 496527 .38313230 Guayaquil 0.4493 541943 .70698084 Sierra Urbana sin Quito 0.7367 437262 .56331707 Sierra Rural -0.4467 585807 .81424722 Costa Urbana sin Guayaquil 0.2085 632177 .75130373 Costa Rural -1.0964 434422 .69612512 Amazonia Urbana 0.2571 44380 .80840630 Amazonia Rural -1.0525 92347 1.05149895 Total 0.1005 3264866 .96987113
56
Appendix 5: Food consumption index
We use the second factor component of this analysis as a control variable in our models. This factor assigns high values to households with high consumption of carbohydrates such as tubers.
Component Matrixa
Component
1 2 3 4 5 6
Total calories consumed on average (parish) .658 -.256 .252 .363 -.132 .130 Carbohydrates from cereal: gr per day consumed on average in every parish .403 -.350 .463 .422 -.149 .362 Carbohydrates from fruit: gr per day consumed on average in every parish .628 .240 -.225 -.121 -.619 -.129 Carbohydrates from milk and derivatives: gr or ml per day consumed on average in every parish .665 -.005 -.465 -.302 .152 .314 Carbohydrates from legumes: gr per day consumed on average in every parish .462 .131 .701 -.502 .056 -.093 Total carbohydrates: gr or ml per day consumed on average in every parish .714 -.071 .412 .342 -.230 .263 Carbohydrates from tubers: gr per day consumed on average in every parish .124 .908 .054 .189 .107 .183 Carbohydrates from vegetables: gr per day consumed on average in every parish .766 .025 -.032 .297 .341 -.369 Fat from meats and derivatives: gr per day consumed on average in every parish .779 -.029 -.143 .005 .061 -.003 Fat from fruit: gr per day consumed on average in every parish .746 .160 -.314 -.046 -.420 -.152 Fat from milk and derivatives: gr per day consumed on average in every parish .691 -.109 -.390 -.406 .207 .333 Fat from fats and oils: gr per day consumed on average in every parish .268 -.337 .531 .156 -.029 .101 Fat from legumes: gr per day consumed on average in every parish .431 .225 .702 -.469 .015 -.117 Fat from tubers: gr per day consumed on average in every parish .061 .937 .042 .223 .112 .188 Fat from vegetables: gr per day consumed on average in every parish .808 .049 -.117 .247 .312 -.342 Protein from meats and derivatives: gr per day consumed on average in every parish .779 .024 -.163 .030 .092 -.008 Protein from fruit: gr per day consumed on average in every parish .725 .209 -.282 -.088 -.551 -.152 Protein from milk and derivatives: gr or ml per day consumed on average in every parish .686 -.135 -.366 -.411 .209 .325 Protein from legumes: gr per day consumed on average in every parish .453 .205 .715 -.459 .065 -.088 Protein from fish and seafood: gr per day consumed on average in every parish .385 -.551 -.002 .100 .147 .060 Total protein: gr per day consumed on average in every parish .915 -.112 .101 .183 .015 .135 Protein from tubers: gr per day consumed on average in every parish .042 .928 .037 .226 .110 .184 Protein from vegetables: gr per day consumed on average in every parish .832 .001 -.091 .215 .293 -.349
Extraction Method: Principal Component Analysis. a. 6 components extracted.
Total Variance Explained
Component
Initial Eigenvalues Extraction Sums of Squared Loadings
Total % of Variance Cumulative % Total % of Variance Cumulative %
1 8.797 38.248 38.248 8.797 38.248 38.248 2 3.458 15.035 53.283 3.458 15.035 53.283 3 3.041 13.220 66.504 3.041 13.220 66.504 4 1.960 8.524 75.028 1.960 8.524 75.028 5 1.443 6.275 81.303 1.443 6.275 81.303 6 1.134 4.932 86.235 1.134 4.932 86.235
Extraction Method: Principal Component Analysis.
Results of principal component analysis of food consumption index Sub-region Mean N Std. Deviation
Quito 0.5279 424982 .0000000 Guayaquil -1.0238 541943 .0000000 Urban highlands 0.4880 594103 1.0451720 Rural highlands 1.1804 489298 1.8296341 Urban coast -1.4320 759663 .4421891 Rural coast -1.8843 306937 .7794456 Urban Amazon -0.3765 56187 .6162099 Rural Amazon -0.3728 67627 .7688065 Total National -0.3627 3240740 1.4087608
List of food items and their food groups
57
Food staple Category
Rice Cereals Barley rice Cereals Oatmeal Cereals Pasta Cereals Cookies Cereals Bean flower Legumes Corn flower Cereals Banana flower Fruits Wheat flower Cereals Machica Cereals Corn y morocho Cereals Mote Cereals Bread Cereals Quinua Cereals Lamb meat Meats and derivatives Pork Meats and derivatives Beef Meats and derivatives Cow entrails Meats and derivatives Chicken Meats and derivatives Chicken piece Meats and derivatives Chicken entrails Meats and derivatives Sausage Meats and derivatives Ham Meats and derivatives Mortadela Meats and derivatives Wiener Meats and derivatives Fresh fish Fish and seafood Tuna or sardines Fish and seafood Shrimp Fish and seafood Clam Fish and seafood Chicken egg Eggs and derivatives Powder milk Milk and derivatives Liquid milk Milk and derivatives Formula (baby milk) Milk and derivatives Cheese Milk and derivatives Yogurt Milk and derivatives Vegetable oil Fats and oils Pig fat Fats and oils Vegetable butter Fats and oils Margarine Fats and oils Butter Fats and oils Avocado Fats and oils Banana Fruits Lemon Fruits Mandarin Fruits Apple Fruits Passion fruit Fruits Melon Fruits Blackberry Fruits Orange Fruits Naranjilla Fruits
Food staple Category
Papaya Fruits Pineapple Fruits Sweet plantain Fruits Plantain Fruits Watermelon Fruits Tomate de árbol Fruits Grape Fruits Melloco/olluco Tubers Potato Tubers Beet Vegetables Yucca Tubers Carrot Vegetables Chard Vegetables Garlic Vegetables Fresh pea Legumes Celery Vegetables Broccoli Vegetables White onion Vegetables Red onion Vegetables Corn in grain Cereals Cabbage Vegetables Cauliflower Vegetables Cilantro and parsley Vegetables Red beans Legumes Brown beans Legumes Lettuce Vegetables Pickle Vegetables Pepper Vegetables Radish Vegetables Tomato Vegetables Pepper Vegetables Dry pea Legumes Corn on cob Legumes Dry red beans Legumes Dry chickpea Legumes Dry brown bean Legumes Lentil Legumes Sugar Sugars Cocoa Sugars Chocolate Fats and oils Brown sugar Sugars Breakfast cereal Cereals Condiments Miscellaneous Salt Miscellaneous Coffee Miscellaneous Water Miscellaneous Mineral water Miscellaneous Powder juice Sugars Juice from concentrate Sugars Soft drinks Sugars
58
Appendix 6: Agricultural Index
Component Matrixa
Component
1 2 3
Gini of land -.689 .204 .055 Productivity of land -.456 .300 .301 Productivity of labour .314 .741 .368 Distance to the highway .765 -.304 .029 % irrigation -.152 .701 .280 % non-agricultural income -.736 -.453 .434 % production for export .805 .116 .186 Technical efficiency .112 .782 .389 Average size of plot .679 -.353 .463 % agricultural income .736 .453 -.434 % land in production -.376 .592 -.559
Extraction Method: Principal Component Analysis. a. 3 components extracted.
Total Variance Explained
Component
Initial Eigenvalues Extraction Sums of Squared Loadings
Total % of Variance Cumulative % Total % of Variance Cumulative %
1 3.736 33.960 33.960 3.736 33.960 33.960 2 2.775 25.228 59.188 2.775 25.228 59.188 3 1.398 12.712 71.900 1.398 12.712 71.900 4 .773 7.031 78.931 5 .654 5.945 84.876 6 .440 4.000 88.875 7 .415 3.773 92.648 8 .325 2.958 95.606 9 .264 2.403 98.009 10 .219 1.991 100.000 11 5.409E-16 4.917E-15 100.000
Extraction Method: Principal Component Analysis.
Report
Subregion Mean N Std. Deviation
Quito -1.5782079 14 .00000000 Guayaquil .8948077 15 .00000000 Urban Highlands -.8920704 400 .83631001 Rural Highlands -.6564539 3727 .83094281 Urban Coast Rural Coast
.4354432
.6389138 659
1875 .55203598 .46746775
Urban Amazon 1.0016837 106 .37435482 Rural Amazon .9985109 1223 .53834390 Total National .0000000 8020 1.00000000
59
60
Appendix 7: Regression results including rural amazon
Table 9 The whole country: Provincial level models
OLS MLM LMLM
(1)
(2)
(3) Raw
(4) Margin Inequality and income
-13.949** (4.45) -14.888*** (3.92) 20.96* (9.284) 3.41* (1.51) Gini Gini2 16.685** (5.37) 17.471*** (4.44) -23.70* (11.11) -3.85* (1.80) H. Consumption 0.183*** (0.04) 0.166*** (0.03) -0.327*** (0.0798) -0.053*** (0.013) Poverty 0.571 (0.50) 0.559 (0.41) -1.110 (1.028) -0.180 (0.167) Mean consumption -0.001 (0.00) 0.000 (0.00) -0.00230 (0.00433) -0.00037 (0.0007) Ln(GDP) 0.101*** (0.02) 0.082*** (0.01) -0.163*** (0.0392) -0.0266*** (0.006) Characteristics of the child D. female 0.149*** (0.03) 0.160*** (0.02) -0.316*** (0.0700) -0.051*** (0.011) Age of child in months -0.143*** (0.02) -0.155*** (0.02) 0.349*** (0.0359) 0.057*** (0.006) Age of child in months2 0.004*** (0.00) 0.004*** (0.00) -0.00993*** (0.00117) -0.002*** (0.0002) Age of child in months3 -0.000*** (0.00) -0.000*** (0.00) 0.0000849*** (0.0000116) 0.00001*** (0.0000019) Proportion of vaccines received 0.140 (0.10) 0.145* (0.07) -0.282 (0.195) -0.046 (0.032) N. months of breastfeeding -0.002 (0.01) -0.003 (0.01) -0.0000995 (0.0146) -0.000016 (0.002) Low birth weight -0.600*** (0.10) -0.612*** (0.06) 1.226*** (0.200) 0.1995*** (0.032) Characteristics of the mother Schooling of the mother 0.026*** (0.01) 0.028*** (0.01) -0.0531*** (0.0125) -0.009*** (0.002) Age of the mother 0.009*** (0.00) 0.009*** (0.00) -0.0217*** (0.00542) -0.004*** (0.0009) Fertility of mother -0.886** (0.31) -0.709* (0.29) 2.046** (0.622) 0.333** (0.101) D. caesarian section 0.108** (0.04) 0.104** (0.03) -0.202* (0.0921) -0.0329* (0.015) D. obstetrician/physician birth 0.063 (0.05) 0.098 (0.05) -0.273** (0.0963) -0.044** (0.016) D. mother underemployed -0.067 (0.03) -0.084** (0.03) 0.126+ (0.0720) 0.021+ (0.012) Characteristics of the household Housing conditions index 0.081** (0.03) 0.099** (0.03) -0.141* (0.0625) -0.023* (0.01) Work experience of head of household 0.002* (0.00) 0.003 (0.00) -0.00858** (0.00285) -0.001** (0.0005) N. children < 12 in household 1/2 -0.148** (0.05) -0.173*** (0.05) 0.312** (0.118) 0.051** (0.019) Other contextual variables R.A. higher education census sector 0.361*** (0.10) 0.296*** (0.07) -0.499* (0.229) -0.081* (0.037) Agro Index 0.172*** (0.02) 0.179*** (0.02) -0.252*** (0.0525) -0.0411*** (0.009) Food consumption index -0.036* (0.02) -0.045* (0.02) 0.0838** (0.0315) 0.014** (0.005) D. rural -0.202* (0.08) -0.122 (0.08) 0.119 (0.175) 0.0193 (0.03) Fixed effects for ethnicity . . .
Indigenous -0.174** (0.06) -0.169*** (0.05) 0.221* (0.108) 0.038* (0.02) Afro-Ecuadorian 0.221** (0.07) 0.246*** (0.05) -0.730*** (0.164) -0.107*** (0.02) _cons -0.354 (1.06) 0.327 (0.88)
sd(_cons)
0.0366 (0.0499) 0.0736* (0.0662) 0.0736 (0.0662) sd(residual)
1.005 (0.0162)
r2 0.283 N 5102
5102
5102 5102 5102 5102
Standardized Beta coefficients for numerical variables and unstandardized coefficients for dummy variables and constant. Dependent variable: ln(z-score of height for age)
61
Table 10 The whole country: County Level Regression Models
OLS MLM LMLM
(1) (2) (3) Raw (4) Margin
Inequality and income -19.402*** (5.00) -15.083** (5.71) 22.20* (11.11) 3.615* (1.809) Gini
Gini2 23.067*** (6.13) 17.567* (6.89) -25.38+ (13.55) -4.13+ (2.206) H. Consumption 0.182*** (0.04) 0.169*** (0.03) -0.320*** (0.0810) -0.05*** (0.013) Poverty 2.196 (1.16) 0.683 (1.58) -1.606 (2.995) -0.262 (0.486) Poverty2 -1.836* (0.76) -0.556 (1.01) 1.432 (1.836) 0.233 (0.299) Mean consumption 0.002 (0.00) -0.001 (0.00) -0.00096 (0.00725) -0.00016 (0.001) Ln(GDP) 0.032* (0.01) 0.042* (0.02) -0.0775* (0.0311) -0.01* (0.005) Characteristics of the child D. female 0.154*** (0.03) 0.157*** (0.03) -0.316*** (0.0705) -0.051*** (0.011) Age of child in months -0.142*** (0.02) -0.157*** (0.02) 0.343*** (0.0361) 0.056*** (0.006) Age of child in months2 0.004*** (0.00) 0.004*** (0.00) -0.00976*** (0.00118) -0.0016*** (0.0002) Age of child in months3 -0.000*** (0.00) -0.000*** (0.00) 0.0000833*** (0.0000116) 0.000014*** (0.00000187) Proportion of vaccines received 0.110 (0.10) 0.165 (0.10) -0.236 (0.196) -0.039 (0.032) N. months of breastfeeding -0.003 (0.01) -0.006 (0.01) 0.00393 (0.0147) 0.0006 (0.002) Low birth weight -0.590*** (0.10) -0.604*** (0.06) 1.222*** (0.203) 0.1991*** (0.033) Characteristics of the mother Schooling of the mother 0.025*** (0.01) 0.027*** (0.01) -0.0521*** (0.0126) -0.008*** (0.002) Age of the mother 0.009*** (0.00) 0.010*** (0.00) -0.0216*** (0.00546) -0.004*** (0.001) Fertility of mother -0.945** (0.31) -0.656* (0.29) 2.069*** (0.628) 0.337*** (0.102) D. caesarian section 0.128*** (0.04) 0.092** (0.04) -0.211* (0.0933) -0.03* (0.015) D. obstetrician/physician birth 0.062 (0.05) 0.089* (0.04) -0.263** (0.0971) -0.04** (0.016) D. mother underemployed -0.061 (0.03) -0.092** (0.03) 0.139+ (0.0728) 0.02+ (0.012) Characteristics of the household Housing conditions index 0.062* (0.03) 0.099** (0.03) -0.112+ (0.0633) -0.0183+ (0.010) Work experience of head of household 0.002 (0.00) 0.003** (0.00) -0.00900** (0.00288) -0.001** (0.001) N. children < 12 in household 1/2 -0.153** (0.05) -0.172*** (0.05) 0.319** (0.119) 0.052** (0.019) Other contextual variables R.A. higher education census sector 0.356*** (0.10) 0.396*** (0.09) -0.587* (0.237) -0.096* (0.039) Agro Index 0.198*** (0.02) 0.226*** (0.03) -0.288*** (0.0631) -0.05*** (0.01) Food consumption index -0.035* (0.02) -0.037 (0.02) 0.0911** (0.0339) 0.015** (0.006) D. rural -0.118 (0.07) -0.055 (0.10) -0.0589 (0.167) -0.0096 (0.027) Fixed effects for ethnicity .
.
.
Indigenous -0.139* (0.06) -0.166** (0.05) 0.202+ (0.113) 0.035+ (0.02) Afro-Ecuadorian 0.211** (0.07) 0.248*** (0.06) -0.728*** (0.167) -0.107*** (0.02) _cons 1.807 (1.13) 1.439 (1.31)
sd(_cons)
0.171 (0.0308) 0.219 (0.0652) 0.219 (0.0652) sd(residual)
0.995 (0.0173)
r2 0.277 N 5096
5096
5096
5096
Standardized Beta coefficients for numerical variables and unstandardized coefficients for dummy variables and constant. Dependent variable: ln(z-score of height for age)
62
Table 11 The whole country: Parrish Level Regression Models
OLS MLM LMLM
(1) (2) (3) Raw (4) Margin
Inequality and income -15.128** (4.72) -12.032* (5.18) 22.01* (10.90) 3.602* (1.78) Gini
Gini2 18.493** (5.86) 14.191* (6.33) -26.79* (13.46) -4.38* (2.20) H. Consumption 0.188*** (0.04) 0.180*** (0.03) -0.351*** (0.0828) -0.06*** (0.013) Poverty -0.456 (0.33) -0.465 (0.39) 1.782* (0.805) 0.292* (0.132) Mean consumption -0.003* (0.00) -0.003 (0.00) 0.00915* (0.00401) 0.001* (0.0007) Ln(GDP) 0.032* (0.01) 0.037* (0.02) -0.0694* (0.0313) -0.01* (0.0051) Characteristics of the child D. female 0.154*** (0.03) 0.154*** (0.03) -0.324*** (0.0713) -0.053*** (0.012) Age of child in months -0.141*** (0.02) -0.156*** (0.02) 0.344*** (0.0363) 0.07*** (0.006) Age of child in months2 0.004*** (0.00) 0.004*** (0.00) -0.00978*** (0.00119) -0.002*** (0.0002) Age of child in months3 -0.000*** (0.00) -0.000*** (0.00) 0.0000834*** (0.0000117) 0.00001*** (0.0000019) Proportion of vaccines received 0.101 (0.10) 0.156 (0.10) -0.235 (0.199) -0.038 (0.033) N. months of breastfeeding -0.003 (0.01) -0.006 (0.01) 0.00433 (0.0149) 0.0007 (0.002) Low birth weight -0.591*** (0.10) -0.599*** (0.08) 1.252*** (0.205) 0.204*** (0.033) Characteristics of the mother Schooling of the mother 0.024*** (0.01) 0.027*** (0.01) -0.0515*** (0.0128) -0.008*** (0.0021) Age of the mother 0.009*** (0.00) 0.010*** (0.00) -0.0217*** (0.00554) -0.004*** (0.001) Fertility of mother -0.955** (0.31) -0.733* (0.31) 2.151*** (0.638) 0.352*** (0.104) D. caesarian section 0.129*** (0.04) 0.086** (0.03) -0.193* (0.0944) -0.031* (0.015) D. obstetrician/physician birth 0.076 (0.05) 0.080 (0.05) -0.247* (0.1000) -0.040* (0.016) D. mother underemployed -0.065 (0.03) -0.090** (0.03) 0.148* (0.0739) 0.024* (0.0121) Characteristics of the household Housing conditions index 0.060 (0.03) 0.108*** (0.03) -0.0999 (0.0653) -0.016 (0.011) Work experience of head of household 0.002 (0.00) 0.003** (0.00) -0.00923** (0.00292) -0.002** (0.0005) N. children < 12 in household 1/2 -0.153** (0.05) -0.161** (0.05) 0.304* (0.121) 0.05* (0.02) Other contextual variables R.A. higher education census sector 0.351*** (0.10) 0.389*** (0.09) -0.640** (0.244) -0.10** (0.04) Agro Index 0.176*** (0.02) 0.238*** (0.03) -0.290*** (0.0643) -0.05*** (0.010) Food consumption index -0.057*** (0.02) -0.041* (0.02) 0.104** (0.0339) 0.017** (0.006) D. rural -0.090 (0.07) 0.010 (0.08) -0.114 (0.153) -0.019 (0.025) Fixed effects for ethnicity .
.
.
Indigenous -0.152* (0.06) -0.165** (0.05) 0.219+ (0.117) 0.038+ (0.021) Afro-Ecuadorian 0.214** (0.07) 0.259*** (0.05) -0.747*** (0.169) -0.11*** (0.022) _cons 2.074* (1.03) 1.416 (1.12)
sd(_cons)
-1.454*** (0.13) 0.347 (0.0707) 0.347 (0.0707) sd(residual)
-0.015 (0.02)
r2 0.276 N 5096
5096
5096
5096
Top Related