3.1 Descriptive Statistics
According to the descriptive analysis presented in Table 1, the mean scores for access to improved sources of drinking water and improved sanitation were 0.279 and 0.194, respectively, in 2000. While the mean score for access to an improved source of drinking increased to 0.619 the mean score for improved sanitation declined to 0.172 in 2016. In line with this, the DHS final report of the country has also confirmed that the sanitation coverage of the country has shown a declining trend. While it can be challenging to identify the precise causes for the decline, they are probably a result of a number of variables such as poverty; the pace of population increase outweighs the intervention/effort from the government to increase sanitation coverage and a lack of resources. Differences in sample sizes, households selected, enumeration areas and other related characteristics between the two survey series may also contribute to the decline in the mean of sanitation coverage.
Other demographic and socioeconomic determinants such as sex of the child, age of the child in the month, age of the mother at birth, sex of the head of the household, and area of residence appear to have a direct or indirect effect on the child’s nutritional status are also included in the analysis. Finally, nine regional states and two city governments are included to capture whether there are regional differences.
3.2 Trends and changes of inequalities in under-nutrition indicators
As we can see from table 2, there has been a sharp and significant decline in both stunting and underweight rates, but the improvement in wasting in children under five has been very slow and stagnant. The table summarizes the absolute gap and relative inequality in child undernutrition across selected socioeconomic covariates and changes in these indicators from 2000 to 2016.
Across all variables, there has been a decline in the prevalence of stunting and under- weight. For example: nationally, the prevalence of stunting and underweight decreased by approximately 19 percent and 17.4 percent, respectively, between the first and last DHS. This is a major achievement in terms of improvements; however, these rates are still higher and require further action to protect children from persistent malnutrition, which in turn affects them later in life. In addition, the absolute differences in the prevalence of stunting and underweight across categories are positive, indicating that disease-related variables are higher in the hypothesized disadvantaged groups (for example a child whose mother is uneducated, who lives in rural areas, has low SES, heads a female-headed household, and has an unimproved source of drinking water).
Table 2 demonstrate the extent of child malnutrition from 2000 to 2016 Demographic and Health Surveys across quintiles, ordered from poorest to richest. The prevalence of both stunting and underweight was relatively higher among children whose families ranked lower. As discussed in the previous section, both the concentration index and concentration curves have been criticized because they say nothing about the middle groups [63]. Instead, they emphasize the two extremes: the privileged and unprivileged individuals or groups. As can be seen in the charts, the relative disparity between the poorest and the richest in stunting prevalence in 2016 is visible.
Similarly, the differences in the prevalence of underweight can be observed between socioeconomic groups in both 2000 and 2016. In both cases, the forms of child undernutrition are higher in the socioeconomically disadvantaged groups than in the comparison group.
Some of the improvements in child malnutrition indicators are in favor of advantaged groups, leading to an increase in inequality between groups, while others are in favor of disadvantaged groups, leading to a decrease in inequality. For example, the absolute difference in the prevalence of stunting among children living in rural and urban areas was about 11.2 percent in 2000, and the difference has increased to 13.7 percent in 2016. This means that the gap has widened and a child living in the city has benefited more than a child living in the countryside.
In contrast, the gap between rural and urban underweight rates has decreased from 14 percent to 11 percent. Similarly, the absolute and relative differences in both forms of child undernutrition by gender of the household head have narrowed. The table shows that children from households headed by men are less likely to be affected by stunting and underweight than children from households headed by women. Both the prevalence of stunting and underweight have improved significantly among children from the poorest and richest nodes. However, the difference between the richest and poorest increased from 11.6 percent to 19.6 percent for stunting prevalence and from 13.2 percent to 16 percent for underweight prevalence within the reported period. In summary, absolute differences in the prevalence of stunting and underweight have not consistently decreased across household wealth quintiles and other demographic and socioeconomic indicators.
The different concentration indices of child undernutrition variables are summarized in table 3. The type of concentration index used in this study is the standard concentration index supplemented by the generalized concentration index. However, Wagstaff Index (W) and Erreygers Index (E) are also obtained for the indicators. The purpose of estimating these indices is simply to support the results of the absolute and relative (standard) and absolute (generalized) concentration indices. It is shown that all types of CIs for the prevalence of stunting, wasting, and underweight in 2000 and 2016 are negative and statistically significant. These estimates showed that the burden is concentrated among the unprivileged groups of children whose families have lower socioeconomic exposure.
As is depicted in the figure absolute concentration curve for the indicators of child malnutrition is above the line of equality and this confirms that the problem is heavily concentrated among children from the lower wealth quintile. These curves only show whether socioeconomic inequalities in health (disease) outcomes exist, or whether inequality is more pronounced at one time than another, in a group, region, or country. It does not generate an estimate of the level of inequality that we use to make comparisons between individuals, groups, regions, or countries.
According to [45] guide to estimating the concentration index, inequality indices can be estimated for binary or categorical variables and compared across different groups. In this study, inequality indices are estimated across different demographic and socioeconomic indicators (rural/urban, educated/uneducated mothers, wealth quintiles, mother’s age at birth, child’s sex, and household heads) and tested with the null hypothesis of no differences between groups. This is done by comparing wealth-related differences in child malnutrition indicators (prevalence of stunting, and underweight) between the above groups.
Since individuals are ranked by their socioeconomic status (proxied by asset index), a negative concentration index indicates that the ill health variable which is malnutrition is highly concentrated among the poor. Given that we have negative concentration indices according to the table presented above, inequalities in malnutrition indicators are said to be increasing, if the value of the index is close to -1 compared to its counterpart. For instance: according to our estimation in table 4, the concentration indices of stunting prevalence have shown an increasing trend at all levels be at national, urban, and rural levels. The index at the national level was − 0.0314 in 2000 and increased to -0.09587 in 2016, in urban areas the index increased from − 0.147 in 2000 to -0.303 in 2016. Finally, the concentration index in rural areas was − 0.014 and − 0.060 in 2000 and 2016 respectively. Those indices demonstrated that there exists a noticeable worsening in the level of inequalities in child malnutrition.
The changes in the concentration index of the indicators roughly confirmed that the changes in the inequality of the indicators from 2000 to 2016 are relatively higher in rural than urban. The estimated standard concentration index is higher for children living in rural than urban. This is due to the reason that households in rural are more likely to own no durable asset, have similar access to utilities and infrastructure (sanitation facility and sources of water for drinking), similar housing characteristics (including the number of rooms for sleeping and building materials) or own very similar properties and categorization of households into quintiles does not clearly show very significant differences in households using the estimated asset index.
As is discussed in the previous sections, PCA generates a well-informing asset index, when the asset is unequally distributed between households [41]. In other words, if an asset is owned by all households or is not owned by any household, it will have insignificant use in differentiating households using their SES. Households who are living in urban in contrast, have a significant difference in terms of the durable asset they own, and the establishment of an asset index based on the PCA can show us very important variance between the richest and poorest households.
3.3 Results of the Empirical Analysis
In the preceding sections, it is attempted to discuss the findings of descriptive analysis and standard concentration indices of inequalities in children’s malnutrition forms across different groups. The results of descriptive analysis and concentration indices generated mixed trends regarding the disparities in children’s malnutrition: that is, their results demonstrated both widening and contracting disparities. However, those results might not vigorous and may be subject to biasedness. A study by [20] concluded that the number of pieces of literature investigating the impact of socioeconomic inequalities on health outcomes and differences in the level of ill-health outcomes among individuals and groups is growing but the explanation about the degree of the differences and explanation of the differences in the inequalities is not sufficiently documented. In the succeeding sections, a discussion of the results of multivariate analysis is presented. In case the dependent variable is a categorical (binary) variable, OLS regression, marginal effects from probit analysis, GLM, LPM, and non-linear logit model can be used to undertake the analysis including the decomposition analysis [60]. As a result, we have employed the Linear Probability model, logit, and probit models where child ill-health variables are predicted by a set of determinants. The purpose of employing all those techniques is simply not beyond checking their consistency. Those techniques have generated comparable results and the coefficients of the determinants will be used in decomposing and estimating the absolute and percentage contribution of the determinants to the socioeconomic inequalities in child health. Studies on disparities in children’s malnutrition are used to establish a relationship between the malnutrition forms and the wealth of their families as well as demographic and socioeconomic experiences. As a result, the main emphasis of the analysis will be given on how the convenient socioeconomic indicators (household socioeconomic rank based on the estimated asset index and mother’s educational status) are contributing towards the socioeconomic disparities in child malnutrition forms.
Table 5 recapitulates the multivariate analysis of inequality in the variable child undernutrition (which is the prevalence of stunting) explained by various inequalities in health determinants. We ran LPM, logit, and probit models for the variable stunting (= 1 if stunted, 0 otherwise) on the Xk determinants, and the models generated comparable results. The coefficients from the LPM with OLS estimates are easy to interpret because they are the marginal effects. However, the interpretation of the coefficients from the probit and logit is not straightforward. The sign of the coefficients from the latter models explains the likelihood that children are stunted relative to the reference groups. Negative (positive) coefficients indicate that the target groups have a lower (higher) probability of stunted compared to the reference groups. Interpreting the magnitude of the probability of being Stunted is more complicated than the estimates from OLS. With LPM, the coefficients of the independent variables are the magnitude of the marginal effect of the covariates on our dependent variable.
According to the regression results, the prevalence of stunting and being underweight are inversely associated with the level of household socioeconomic status. From the results of LPM, children from relatively high SES are less likely to be stunted than their peers. A child from a poorer, middle, richer, and richest household is 3.3 percent, 9 percent, 11 percent, and 14.4 percent less likely to be stunted than a child from the poorest household, respectively (although the coefficient for the poorer household is not statistically significant), after controlling for all other independent variables. The estimated coefficients also confirmed that probabilities decrease with household SES. In the same way, the results from the logit model are quite similar. According to the results from the logit model, the probability of having stunted children is 3.5 percent, 9.6 percent, 11.5 percent, and 15.4 percent lower from the second to the fifth quintile, respectively. The results based on the DHS 2000 on the influence of SES on the probability of being disabled are not statistically significant.
According to the regression results summarized in Table 6 for the underweight frequency indicator, the probability for a child to suffer from being underweight is strongly related to the socioeconomic status of the household. We confirmed that inequality in SES causes inequality in the probability of a child being undernourished as measured by being underweight. From the model LPM, a child is 3.4 percent, 7.8 percent, 13.4 percent, and 10.7 percent more likely to be underweight if he or she comes from a poorer, middle, richer, and richest household, respectively. The marginal effect of household socioeconomic exposure on the probability of a child being malnourished supports the findings obtained from LPM. Holding other things constant, the odds of a child being malnourished are 3.3 percent, 7.7 percent, 13 percent, and 10.4 percent lower than the poorest in rank when the family’s SES is in the 2nd to 5th percentile.
We have discussed in the previous section that the educational status of a mother plays a very important role in minimizing risks associated with the nutritional status of her children. Mothers with higher educational status make greater use of prenatal care during pregnancy and postnatal care. Wise use of this type of care reduces the likelihood that a child will be malnourished. We categorize mothers with primary education and fewer as uneducated and those mothers who have secondary education and more are educated. The results of the regressions are consistent with these types of findings. Leaving the other covariates unchanged, the probability of being stunted is higher for those children whose mothers are less educated (uneducated). According to the results from LPM and the logit models, the probability of a child being stunted is 16.2 percent and 11.8 percent higher, respectively, if the mother is uneducated.
In terms of the age of a child, the results are statistically significant and confirm our expectations. The results from LPM show that the marginal effect of a child aged 13–36 months and 37–59 months has a higher probability of stunting than a child aged 0–12 months. In 2000, controlling for other covariates, a child aged 13–36 months and 37–59 months is 34.8 percent and 34.2 percent more likely to be stunted than a child aged 0–12 months, respectively. Similarly, the probabilities for these age groups of being stunted decreased to about 27 percent and 25.1 percent, respectively. The same is true for the results from the logit and probit models (which are attached at the end of this document). According to the logit model, a child aged 13–36 months is 27.9 percent more likely to be stunted than a child aged 0–12 months.
For children 37–59 months old, the probability of being stunted is estimated to be 25.8 percent higher than for children 0–12 months old. The results from LPM and logit regressions using the 2016 data give us very comparable probabilities. The results for the underweight indicator are also similar. Daughters/sons who are younger than household members receive more care from household members compared to those who are older. Children aged 13–36 months and 37–59 months have a higher probability of suffering from being underweight than the reference group (children aged 0–12 months). The LPM regression results show that holding all other covariates constant, 13–36 and 37-59-month-old children have a 16.6 percent and 16.3 percent higher probability of suffering from being underweight than the children aged 0–12 months. In 2016, the probability of being underweight for their age was 10.3 percent and 13 percent higher in the above age groups, respectively. Complementing this, the logit model produces very similar probabilities. The probabilities for children aged 13–36 and 37–59 months of being underweight were 10.3 percent and 12.9 percent, respectively, which are very similar coefficients to the LPM regression results.
Other variables such as a child’s birth order, the child’s sex, the head of the household, the source of drinking water, and regional conditions also have a significant impact on child malnutrition disparities. Children who do not have access to an improved source of water for drinking are affected by both stunting and being underweight. The motivation for including regional state and city governments in the analysis was to capture any regional effect and to examine who was doing better than who. According to Table 5, in 2000, the probability of being underweight for age was relatively lower in Afar, Oromia, Somali, Gambela, Harari, Addis Ababa, and Diredawa regions compared to Tigray regional state. In 2016, children from the only Somali region, Gambela region, and Addis Ababa city are less likely to be stunted compared to Tigray regional state. Children from Harar town, Diredawa town, and Addis Ababa town were less likely to be underweight in 2000 than in the reference region. Gumuz region, Addis Ababa town, and Diredawa town performed better than Tigray regional state in 2016.
In conclusion, there is still a significant difference between the regional states in terms of the likelihood of being undernourished. The variable “type of household sanitation” is not statistically significant in all models, so it does not contribute to the inequalities in
child malnutrition. Another important variable that is found to have a significant effect on inequalities in stunting and underweight according to the regression results is the age of the mother. For example, according to the result of LPM, the older the mother is, the lower the probability that a child is smaller than his mother for his age. The estimated probabilities are 5.3 percent, 9.1 percent, and 9.8 percent lower for a child whose mother’s age is 20–29 years, 30–39 years, and 40–49 years, respectively, compared to a mother less than 19 years old.
Disparities in child malnutrition according to the place of residence (urban and rural) produce mixed results in recent studies. There is scattered evidence that the gap is narrowing and that migration contributes to this effect. Poverty and malnutrition are gradually shifting from rural to urban areas in developing countries. Thus, the number of urban poor and malnourished is increasing faster than the number of rural poor. Other studies have shown that urban children are better nourished and less likely to be stunted and underweight than their rural counterparts [51, 21]. According to [51], urban children are less likely to be stunted and underweight because they are better nourished than their rural counterparts. Increased poverty-related migration from rural to urban areas decreases the urban health advantage [28]. These migrated people live in informal settlements and slums, making them vulnerable to disease. All our regression results showed that a child living in rural areas is less likely to be stunted and more likely to be underweight than a child living in urban areas. However, the relationship is not statistically significant.
3.4 Decomposition Analysis
Once we obtained the coefficients of the determinants of health inequality, we can proceed to decompose the contribution of the covariates to inequality in child malnutrition. Tables 7 and 8 show the absolute and percentage contribution of health determinants to inequalities in child undernutrition (stunting and underweight prevalence). The primary aim of these tables is to distinguish the main demographic and socioeconomic determinants of health that contribute to inequalities in poor health indicators at a representative national level.
Based on the data from 2000 Ethiopia DHS, the selected explanatory variables contributes to around 97 percent of the disparities in stunting prevalence, and only 10 percent of the inequality in underweight is explained by the inequalities in the determinant of inequality if underweight. Mothers’ educational status contributes substantially to the disparity with about 31.3 percent followed by socioeconomic status which contributes about 30.4 percent. This indicates that the percentage contribution of inequalities in both variables (education level of mothers and SES) accounts for around 62 percent of the inequality in stunting. Likewise, being born at later orders has also an appreciable contribution to the inequality in stunting. According to the decomposition in table 7, inequalities in mothers’ educational status, socioeconomic status, and child’s birth order have contributed significantly and can be considered the most important determinants of inequality in stunting.
The decomposition of the contribution of inequalities in the determinants of health to the inequalities in stunting and underweight prevalence using the DHS 2016 are summarized below. According to the decomposition analysis, the explanatory variables have contributed to about 92.6 percent and 93.8 percent of the inequalities in stunting and underweight prevalence respectively. The unexplained component of the inequality in stunting is 7.4 percent and 6.2 percent in underweight. Compared to the decomposition analysis discussed above, this one is relatively explained better by the perspective covariates. Both socioeconomic position and educational level of mothers have contributed more than 90 percent to the inequality in stunting and about 80 percent to the disparities in underweight prevalence. The birth order of a child has also a relatively significant contribution to the inequality: those children who are born later orders are more probable to be malnourished. The contribution of birth order accounts for about 4.5 percent of the inequality.
Bringing those tables 7 and 8 together, we can also compare the percentage contribution of the major determinants of health into the disparities in child malnutrition. In 2000, all the corresponding variables have explained the disparities in stunting more than they did in 2016. The percentage contribution of the socioeconomic position of households has increased from 30 percent in 2000 to 77 percent in 2016. However, the percentage contribution of mothers’ education declined from 31 percent in 2000 to 13.5 percent in 2016. Even if there exist fluctuations in the percentage contribution of the important and conventional socioeconomic indicators (mother’s education status and wealth index of households), both remain to be the driving variables for the inequality in child malnutrition in Ethiopia.
Like other poor countries, it is revealed that in Ethiopia there were considerable improvements in dropping off child malnutrition indicators (both stunting and underweight, whereas the changes in wasting prevalence, are not large enough) in the past decades. But those global improvements are not evenly distributed [5, 57, 21] among various socioeconomic groups. The same is true in the case of Ethiopia; despite the improvements in the indicators under study between 2000 and 2016, some of the improvements are in favor of the advantaged groups causing the inequality between groups to increase and few other improvements are in favor of the disadvantaged ones leading the inequality level to be contacted. The absolute gap in stunting prevalence among children living in rural and urban was about 11.2 percent in 2000 and the gap has increased to 13.7 percent in 2016. This implies the gap has increased and a child in an urban has benefited more than a child living in a rural.
In contrast to this, the difference between the underweight rate in rural and urban has dropped from 14 percent to 11 percent. Residence locations and distance to get healthcare services are the most common geographical factors [19, 48] that contributed to higher child health problems. A serious concern is that most child health problems happen from causes that can be easily manageable or preventable. The life chances of children vary dramatically by location and early life experience. A girl born in a poor neighborhood can expect to spend more of her life suffering from health problems than had she been born to rich relatives.
Likewise, the absolute gaps and relative gaps in both forms of children’s malnutrition had contracted by the sex of the household head. Both stunting and underweight prevalence among children from the poorest and richest knots have significantly improved. However, the difference between the richest and poorest in the stunting prevalence has in- creased from 11.6 percent to 19.6 percent and underweight prevalence has increased from
13.2 percent to 16 percent respectively within the specified period. To sum up, the ab- solute gaps in stunting and underweight prevalence have not consistently declined across household wealth quintiles and other demographic and socioeconomic indicators.
The findings from the empirical analysis are statistically significant and consistent with other empirical studies when it comes to a child’s age. According to LPM results, children between the ages of 13 and 36 months and 37 and 59 months are more likely to be stunted than children between the ages of 0 and 12 months. A child’s likelihood of being stunted increases as they become older, according to studies conducted using the data from DHS 2000 and DHS 2016. The likelihood of being stunted dropped for these age groups from 2000 to 2006, indicating that the unfavorable health effects are getting better. The underweight indicator produces similar findings. Daughters and sons who are younger than household members get more attention from the family than those who are older do. Children between the ages of 13 and 36 months and 37 and 59 months are more likely than the reference group to be underweight (children aged 0–12 months)