Shocks in food availability and intra-household resources allocation: evidence on children nutrition outcomes in Ethiopia

Many of the intra-household child gender welfare studies using human capital investment show significant inequality between boys and girls (Quisumbing and Maluccio, 2000; Ejrnæs and Pörtner, 2004; Fafchamps et al. 2009; Behrman et al. 1982). Other welfare outcome studies based on child nutrition use anthropometric indicators as an alternative measurement technique to overcome the absence of child individual expenditure and child productivity information in many datasets.

Findings regarding intra-gender child nutrition inequality within a household show contextual evidence. Studies from South Asia indicate that a girl is worse off than a boy in nutrition outcomes (see for example, Behrman, 1988; Pal, 1999; Dancer et al. 2008) while in most Sub-Saharan African countries children are either equally malnourished or boys have less nutrition than girls (Garret and Ruel, 1999; Linnemayr et al. 2008; Quisumbing, 2003; Svedberg 1990). These findings indicate that child gender is found to be a determining factor in the intra-gender children nutrients allocation (see for example, Koohi-Kamali, 2008).

In the intra-gender resources allocation, the decision is mainly done by parents where child gender preference influence is the reason for the per capita human capital investment and health expenditure inequality between children’s sex within the household. Parents’ child gender preference refers to the attitude that parents give values to their kids on the basis of cultural and religious traditions and community norms, shaping individual attitudes and behavior (Behrman et al. 1982; Hank and Kohler, 2000; Clark, 2000; Rahman and Rao, 2004).

We used the Living Standards Measurement Study-Integrated Survey on Agriculture (LSMS-ISA) collected by Ethiopian Central Statistical Agency and the World Bank in two waves in Ethiopia. The first and second survey rounds were conducted in 2011/2012 and 2013/2014 respectively. We used these two round dataset (i.e., we exclude the third round dataset) for two reasons. One is that we started the work earlier than the third round dataset is freely available. The second reason is the fact that our focus of analysis is on children less than 5 years old; we choose to use the follow-up anthropometric information of a particular child who is in the less than 5 years age range in the two successive rounds.

According to the survey’s Basic Information Document of LSMS survey, the Ethiopian Rural Socioeconomic Survey (ERSS) sample is drawn from a population frame that includes all rural and small-town areas of Ethiopia except for three zones of Afar Region and six zones of Somali Region.

The survey encompasses nine regions and one federal city administration (i.e., Diredawa); 69 zones are randomly drawn from these survey areas where 290 rural EAs² and 43 small-town EAs in total 333 EAs are included. Accordingly, 3969 households are randomly selected from the EAs.

Our interest is an individual-level welfare outcome variable (i.e., anthropometric indicators), as dependent variable, in reference to World Health Organization (WHO) standards used to calculate the z-score of child nutrition under five.

>The z-score, welfare indicator at child level, is the outcome variable taken as a good proxy for child welfare so it is used to examine the welfare sharing among children within the household. We exclude the biologically implausible results of the z-score (dropped z-score results which are greater or equal to the absolute value of 6). For the purpose of examining the behavior of resources/nutrient allocation between genders, from the 3969 households, 1428 children are included in the regression after excluding children who do not have at least a sister or a brother of age below 5 years in the household.

In terms of height-for-age z-score, close to 44% of the boys and 40% of the girls are stunted. Yet, girls slightly fare better in nutrition than boys (see Table 2). Similarly, in Table 2, weight-for-length/height z-score shows that both boys and girls are wasting close to 10%.

According to the World Health Organization (WHO, 1995), if population prevalence of malnutrition (i.e., for % of children < 60 months years old) below – 2z-scores is between 25 and 29%, then the degree of prevalence is high where EDHS report in 2011 confirms the high degree of prevalence in Ethiopia.

Child welfare outcomes are child nutrition measured by the anthropometric indicators, namely, weight-for-age z-score (waz06), length/height-for-age z-score (haz06), and weight-for-length/height z-score (whz06). We used two waves’ panel data, and we report the weight-for-age z-score (waz06) because it is a composite⁶ nutrition outcome indicator.

Child gender dummy is among the individual time-invariant variables of a child used for resource allocation comparison between children in a household.

During food scarcity, parents follow a pure investment strategy, exposing their more vulnerable children to greater malnutrition risk (see for example, Behrman, 1988; Chiappori, 1997; Dercon and Krishnan, 2000; Fafchamps, Kebede and Quisumbing, 2009; Thomas, 1990). Gender variable and its interaction with an aggregate household level shock index variable are included in the estimation to test if child welfare outcomes disparity exists. We used drought, flood, heavy rain, landslide, and crop damage variables to build our shock index⁷ variable. This aggregate environmental shock index is constructed to measure household-level resource-sharing behavior in the preceding season. The dummy responses for each of these shocks represents whether shocks occurred or not; 1 represents the occurrence of shock while 0 represents non-occurrence of shock. The sum of the dummies is averaged to the number of questions asked about the shocks. The occurrence of all the shocks is equal to 1 while the non-occurrence is 1.

In child nutrition analysis, mother’s education level is the common child nutrition predictor. Unlike father’s time,⁸ mother’s education level serves as a proxy for the cost of children because mothers are mainly responsible for rearing a child which is also known as the opportunity cost of their market wage.

Child breastfeeding duration in months, the age of a child in months, and medical aid or aid consultancy are also among the individual specific predictors included in the regression.

Dummy variable if spouses live together in the household is also included to see if a collective agreement on resource allocation differently affects more the welfare of different-sex children than the ones whose spouses do not live together.

We incorporate region variation variable which enables us to compare the child nutrition differences in different regions of the nation. Amhara, Oromia, Tigrai, SNNP,⁹ and “other¹⁰ regions” variables are included in our regression where Amhara Region is the reference category in our factor variable of the regional variation analysis.

Time devoted to child care is one of the excellent predictors of child nutrition and health where the distance to the main road and land/plot to the household competes mothers’ time.

We interpret the anthropometric measurements of nutrition, a good indicator of intra-household resource allocation (Horton, 1988), using weight-for-age z-scores because it is a composite index of height-for-age and weight-for-height. The first is an index showing stunting; an indicator of linear growth retardation and cumulative growth deficits in children. Stunting is the result of failure to receive adequate nutrition over a long period of time and recurrent and chronic illness. The second is an index indicating wasting; this measures body mass in relation to body height or length and shows current- or short-run nutritional status of a child.

As it can be seen in Table 4, male dummy confirms that the weights ratio equality in Equation (8) is stable. The hypothesis for the equal coefficient for the individual time-invariant variable, male dummy, is not rejected. This goes in contradiction with the findings by Svedberg (1990), Christiansen and Alderman (2004), and Peterman et al. (2014) that provide evidence on the presence of parents’ gender bias on resource allocation.

The boy’s nutrition, the standard deviation of nutrition (z-score) as compared to World Health Organization standard reference population, is not affected by our male dummy variable. We find this result very interesting because, unlike the literature findings using the per capita analysis, it shows that the optimal resource allocation is not violated by gender.

Three possible reasons for the ambiguity due to the inconsistency of our result: one might be due to the data structure used for analysis. All of the empirical findings we have seen thus far consider that their data structure is a non-clustered/non-nested dataset. In practice, individual nutrition information is clustered at the individual level while resource allocation decision-making is done at the higher level, household level. That is, the individual-level variable is nested at the higher-level actions, household-level actions, which require hierarchical modeling and hierarchical estimation technique such that it captures the inter-cluster differences between boys and girls. Therefore, our hierarchical model analysis adjusts for unmeasured heterogeneity that exists in panel data studies. Hierarchical model estimation offers substantial benefits over classical, non-hierarchical approaches (Feller and Gelman, 2015).

The second possible reason can also be biological differences in trouble tolerance between boys and girls, i.e., given less than adequate food supply, girls tend to cope better with a shortage of food than boys from the standpoint of their bodily development (see for example, Marcoux, 2002). In the presence of resource gap and allocation disparity between boys and girls, boy’s nutrition is expected to be better off than girls, if girls cannot cope with troubles better than boys. Excellent evidence for the existence of differences between boy and girl groups is the inter-class correlation coefficient¹¹(ICC) of our hierarchical/mixed-effects model. ICC result is in between 0 and 1 (that is, 0.46 between individuals and 0.63 on average) which indicates a difference in variance between boys and girls due to the biological bodily development so it is a reason for nutrition equality while there is allocation disparity.

The third argument is that boys at this age do more physical exercises relative to girls (see for example, Timmons et al. 2007; Timmons et al. 2012). That is, boys spend more energy than girls such that the boys’ body weight during their preschool age makes the boys’ nutrition equal to the girls’ nutrition despite allocation disparity. In Table 5, during bad times, while boys still exercise/spend energy more than girls, the result shows an insignificant negative gender effect on nutrition with an unequal coefficient which hints on the existence of a slight pro-boy resource allocation bias.

Our conclusion is that the equality in nutrition is only due to differences in coping troubles and child’s physical activities.

Now let us turn to other child nutrition predictors. Consistent with the literature on nutrition (Behrman et al. 1982; Behrman, 1988; Horton, 1988), child age indicates an inverse relationship with nutrition; as age increases, a child’s nutrition declines, while breastfeeding duration in months shows a positive correlation with nutrition. The statistically negative significance of child age in months complements the argument that physical exercise can facilitate weight loss and affect nutrition.

Another evidence is the relationship between getting individual medical aid and child nutrition.

Unlike the expectation, these variables have an inverse relationship though statistically insignificantly correlated.

In normal circumstances, individuals who get medical aid and health consultancy are expected to be healthy and have more nutrition than the ones who do not. The explanation for the inverse relationship of these variables here can be the community practice; parents or guardians take children to health centers for medical aid/consulting after they get sick.

A value of 1 in our dummy most probably indicates that the child is suffering from an illness which implies that the child have less nutrition with a probable existence of endogeneity problem between illness and nutrition (Dercon and Krishnan, 2000).

Maternal education may have a bigger impact on child welfare distribution and nutrition. Our result on mothers’ education level, a usual child’s nutrition predictor, confirms that mothers with primary, secondary, and tertiary education level have a more positive significant effect on child’s nutrition as compared to mothers with no education.

Mother’s hours spent in fetching water and other agricultural activities show an insignificant negative effect on child’s nutrition.

Area variation incorporated in our regression shows that a boy from Tigrai Region shows a more insignificant negative coefficient than a boy in Amhara Region (the reference category in our factor variable).

Compared to the Amhara Region, a boy being in Oromia has more nutrition than a boy in Amhara; the evidence is positive significant coefficient at 1% level of significance.

SNNP region variation is not a significant effect on nutrition while the coefficients for other regions¹² show a positive correlation that a boy is better in nutrition than in Amhara Region.

We can conclude that a child who grows in Tigrai, a region located at the most northern part of the nation, usually associated with war front and drought-prone area, is less nourished than a child who grows in any other region of the nation.

Mom’s participation in PSNP shows that an inverse effect on nutrition may be due to the composite income effect earned from different modality of PSNP participation¹³(Quisumbing, 2003). According to Quisumbing (2003), the impact of PNSP on child nutrition depends on whether the household is the recipient of FD or FFW. Our dataset does not have participation modality information; therefore, we remain inconclusive about the result.

As expected, plot distance to the household in kilometers shows significant inverse correlation with nutrition.

A number of rooms owned by the household indicate positive significant correlation with nutrition. Most of the time, a number of rooms and household wealth are positively correlated implying that wealthy people own a large number of residence rooms. This correlation can also imply that children from wealthy households have relatively more nutrition than those from the poor households.

Household size, number of sisters, and number of meals share variables do not have a significant effect on nutrition. One possible reason for this might be the fact that children at this age are not supposed to share the burden of consequences of these variables.

Thus far, we discuss the determining factors of child nutrition with particular emphasis on the impact of parent’s child gender preference on nutrition. Furthermore, the interaction of aggregate household level shock index and male dummy variables gives us a clue about the welfare allocation between genders within the household. In both cases, we tried to see if the unobserved parent’s child gender preference affects a child’s nutrition.

Like the slight median breastfeeding difference between boys and girls in the Ethiopian Demographic Survey (2011) report, the boys’ median breastfeeding, in our case, is slightly greater (25.7 median breastfeeding) than girls’ (24.8 median breastfeeding) which indicates small mean breastfeeding difference between boys and girls.

The t test mean comparison result supports our justification for why resource allocation is biased towards boys although the study is about both boys’ and girls’ nutrition.

Taking breastfeeding duration as a proxy variable for welfare distribution among children in a household, the small mean deviation in breastfeeding duration in a month indicates that the social planner in the household allocates a bit biased towards boys’ welfare weights.

By implication, male children to a female children welfare-weight ratio within the household would remain slightly unequal.

Children growth standards show that children born worldwide should grow in equal growth status regardless of where they were born. Adequate nutrition, environment, and health are the major determinants of child growth where child gender and ethnic origin are minor determinants. Nevertheless, many empirical studies on nutrition indicate that there are deviations when examined against the WHO 2006 growth standards and other standard cut-offs.

Similar to the EDHS (2011) report, this paper, using the Ethiopian LSMS dataset collected in 2011/2012 and 2013/14 confirms that close to half of Ethiopian children up to 5 years old are below the WHO 2006 growth standards. Out of 880 boys, 43% of them are stunted while for 40% of the 548 sample of girls are stunted. Furthermore, 26% of boys are underweight relative to 24% for girls.

Results from the mixed-effect estimation depicted in Tables 4 and 5 show unbiased nutrition outcome with gender. At a glance, our results with and without imposing the aggregate household level shock index variable indicate that there is no nutrition bias between genders.

Empirical evidence based on different measurement techniques, namely, using per capita human capital investment and per capita health expenditure and anthropometric indicators do not show the same result. The per capita expenditure estimations show that resources allocation in a household is biased against girls. On the other hand, the empirical works using anthropometric indicators confirm the contrary. Many of them find that girls have more nutrition than boys. We find inconsistent evidence in this aspect. Our results indicate that boys and girls have equal nutrition. Our conclusion to this contradiction is deduced from the three possible scenarios we have explained above. The nutrition equality is due to differences in stress tolerance and physical exercises between boys and girls. Even though resource allocation is biased against girls, girls’ bodily development enables them to tolerate stress such that their nutrition is equal to boys’ during food shortage. Another scenario might be that, in this age group, boys exercise more than girls (Timmons et al. 2012). Therefore, boys’ physical exercise reduces boys’ body weight (i.e., it negatively affects the anthropometric indicators of a boy), compared to girls’ body weight which in turn makes boys’ have equal nutrition as girls even though boys get more resources than girls.

In bad times, had resource been allocated equally between genders while boys still exercise/lose body weight than girls, nutrition allocation (i.e., the anthropometric indicator) should definitely be in favor of girls. Therefore, in both good and bad time cases, our estimation result confirms that there is nutrition allocation bias against girls.

From a policy perspective, knowing the relative intra-household differences in nutrition for the different groups within the household is important. Therefore, what we have found is an evidence that boys at this age group spent more energy than girls, and girls are better in stress tolerance so all these compensations make nutrition equal while resource allocation is biased. Nutrition inequalities would remain large had resources allocation been not biased towards boys. Such resource allocation differences, while there is seemingly nutrition equality between boys and girls, can lead to short-run drawback (such as difference in vulnerability to disease) on boys’ health and it may also lead to permanent effects and diminished health and education outcomes that may also be a reason for adulthood inequalities in the long run (see for instance, Alderman et al. 2006; Case and Paxson, 2006).

One concern on these findings is the household structure and bargaining power of decision-making process in the household. Bargaining power of decision-makers at the household level, particularly women’s empowerment, plays a crucial role on food consumption allocation. Our next research interest is to see the effect of women’s empowerment on the intra-gender child nutrition disparity within a household.