The research employed a cross-sectional study design utilizing secondary data from the 2019 Ethiopia Demographic and Health Survey (EDHS). The study was conducted in Ethiopia, a country bordered by Sudan to the west, Somalia and Djibouti to the east, Eritrea to the north, and Kenya to the south. Ethiopia covers an area of 1,112,000 square kilometers and is divided into 11 regions, in addition to two municipal governments.

The study focused on children aged 6–23 months in Ethiopia. The data used in this research were sourced from the 2019 EMDHS, specifically extracted from the birth record (BR) dataset available in the DHS Program. Access to this data was granted through personal accounts after justifying the request (Koroso et al., 2021). We extracted dependent and independent variables from the EDHS 2019 data, which includes the full birth history of all children whose mothers were interviewed. The data set contains health and nutrition data for children born in the last 5 years (Rutstein and Rojas, n.d.; Daniel et al., 2012).

The household wealth index, maternal education, and regions of Ethiopia were used for the inequality analysis of the concentration index (CIX).

Data cleaning, recoding, variable generation, and labeling were performed using STATA version 17.0. We applied sample weights to account for unequal probabilities of selection across geographically defined strata and non-responses. Means and standard deviations were calculated for normally distributed continuous variables, following the skewness and kurtosis tests. In addition, frequencies and percentages were computed for categorical variables to describe the characteristics of the study population. Throughout the analysis, the survey design was accounted for and appropriately declared.

The concentration curves (CCs) were plotted, starting with the most disadvantaged groups and ending with the most advantaged groups, to illustrate the cumulative proportion of MDD (y-axis) against the cumulative proportion of the children, sorted by household wealth and geographic region (x-axis). A 45-degree slope, extending from the northwest to the northeast, represents perfect equality. If the CC overlaps with the line of equality, it indicates that MDD was equally distributed among the children aged 6–23 months across the household wealth and geographic regions. However, if the CC deviates above or below the line of equality, it suggests the presence of inequality. A curve below the equality line indicates that MDD was more concentrated among the advantaged groups. The degree of inequality increases as the curve moves further away from the line of equality (O’Donnell et al., 2016).

The purpose of the decomposition was to determine the contribution of different independent variables to inequalities. A chi-squared test (x²) was used to examine statistically significant inequalities in MDD by wealth index and region. Before the decomposition analysis, bi-variable logistic regression was fitted, and variables with p-values less than 0.25 were selected for the decomposition analysis. Prior to conducting the decomposition analysis, multicollinearity was checked among the explanatory variables using the variance inflation factor (VIF), and no multicollinearity was detected since the mean–variance inflation factor (VIF) was 2.28.

To assess inequalities in MDD consumption, we calculated the concentration index, which is a widely used measure of socioeconomic inequality (O’Donnell et al., 2016). This index was calculated based on the following formula (Equation 1): (1) C I = 2 μ c o v h r

where CI is the concentration index, r is a person’s fractional rank in the distribution of socioeconomic position (Wagstaff et al., 2007), μ is the mean of the health variable, h is the health variable measured for inequality, and cov indicates the covariance. The value of the concentration index ranges from −1 to +1. A negative value indicates a disproportionate concentration of MDD consumption among the disadvantaged groups, whereas a positive value indicates a disproportionate concentration of MDD among the advantaged groups.

As our outcome was a binary variable, the bounds of the concentration index did not extend to −1 and +1 but could equal to μ. Therefore, we normalized the concentration index by dividing its value by its bound, as proposed by Wagstaff (2005, 2011) (Equation 2). (2) Cnom = C 1 − μ

We broke down the concentration index to calculate the explanatory variables’ relative contributions to wealth disparities in MDD, aiming to identify the variables causing the observed disparities in dietary diversity. For a set of determinants (Xk) and MDD (y) in a linear additive relationship, such as (Equation 3): (3) y = α + ∑ k β k χ k + ε

Here, 𝛽_𝑘 is the coefficient of 𝑥_𝑘, and 𝜀 is the error term. The concentration index of y denoted (C) could be written as follows (Equation 4): (4) C = ∑ β k x − k μ C k + G C £ μ

In this context, μ is the mean of y, ⁻xₖ is the mean of xₖ, and Cₖ is the concentration index for xₖ, defined similarly to C. In addition, βₖ ⁻xₖ = μ indicates the elasticity of MDD with respect to the explanatory variables, while G C_∊/μ represents the generalized concentration index for the error term ∊ (O’Donnell et al., 2016).

A negative contribution revealed that an independent variable contributed to a pro-poor distribution of MDD. In contrast, a positive contribution indicated that an independent variable contributed to a pro-rich distribution (Equation 5). (5) Cnorm = ∑ k β k X k / μ C k 1 − μ + G C ε / μ 1 − μ

We employed a generalized linear model (GLM) with a binomial family and a logit link to decompose inequalities in MDD because the outcome variable was binary.

From the 3,145 weighted samples of mothers with children aged 6–23 months, the mean age and standard deviation values of the mothers and the children were 33 ± 5 years and 11.09 ± 4 months, respectively. Regarding the consumption of MDD among the children, 8.5% of the children with mothers aged 35–49 years, 35.2% of the children with married mothers, 26.2% of the children with mothers who had a higher education level, and 15.1% of the children with mothers having a wealth status consumed adequate MDD (Table 1).

Approximately 6.2% of the male children, 7.2% of the children aged 6–12 months, and 35% of the children with birth intervals greater than 24 months in Ethiopia consumed adequate MDD (Table 2).

The Concentration index (CIX) for MDD by wealth status was 0.3, with a p-value of less than 0.05, indicating that the consumption was concentrated among the children in the higher wealth status group (pro-rich).

The CIX for MDD by maternal educational status was 0.085, and the p-value of 0.087 suggested that there was no statistically significant inequality.

The CIX for MDD by the geographic regions of Ethiopia was 0.084, and the p-value (<0.05) indicated that MDD consumption among the children was significantly concentrated among the respondents from the developed and urban areas, as the positive CIX suggested (Table 3).

The prevalence of adequate MDD in Ethiopia was 7.11%. The concentration curve of MDD against wealth status showed that the curve was below the 45° line, indicating that MDD consumption among the children was more concentrated among the mothers with a higher wealth status. In the graph, the X-axis represents the percentage of the children with varying levels of dietary diversity by wealth status, the Y-axis displays the cumulative proportion of the children achieving specific dietary diversity levels, and the L (p) line indicates the cumulative proportion of the children meeting the MDD criteria.

It shows how MDD consumption was distributed among the children based on their family’s wealth status. The shaded region around the L (p) line illustrates the 95% confidence interval (CI) of the estimates, indicating the range within which the true values are likely to fall.

The curve was generally found to rise with increasing wealth status, indicating that the children from the wealthier families tended to have higher MDD consumption (Figure 1).

The concentration curve of MDD against the regions of Ethiopia was below the 45° line, indicating that MDD consumption among the children was more concentrated among the mothers from the developed and urban regions of Ethiopia. The curve’s shape suggests that there were regional differences in MDD consumption, with the developed regions showing a higher cumulative proportion of the children achieving a diverse diet compared to less developed regions (Figure 2).

The study, as presented in Tables 4, 5, showed the decomposition of the concentration index (CIX) for MDD consumption according to the wealth status and regions of Ethiopia, respectively. A negative percentage contribution indicated that the factors helped reduce inequality, while a positive percentage contribution indicated that the variables helped increase the observed inequalities.

Regarding maternal educational status (0.88%), elementary education and secondary education were found to have negative contributions, while higher education had a significant positive contribution to MDD consumption. The sex of the household head (0.3%) and region (3.1%) contributed to increasing the observed inequalities in MDD consumption between the wealthy and poor regions (Table 4). Wealth status (11.06%), maternal age (6.7%), maternal educational status (1.87%), and place of residence (0.016%) contributed to increasing the observed inequalities in MDD consumption in the less developed and developed regions of Ethiopia (Table 5).

The concentration of adequate MDD consumption among the children in the wealthy, developed, and urban regions indicates the need for focused efforts targeting disadvantaged groups in poor, rural, and less developed regions.

The findings call for policy actions to reduce poverty and improve maternal education levels, as these were major contributors to wealth-based and geographic inequalities in MDD. National policies should ensure equitable distribution of resources and access to diverse foods across all regions of Ethiopia.

This study’s robustness lies in its use of nationally representative data, enabling generalizability to all Ethiopian children. In addition, the application of a decomposed concentration index approach allowed for a detailed analysis of socioeconomic and regional inequalities in MDD consumption, providing valuable insights into the contributing factors.

The study’s cross-sectional nature limited its ability to establish causal relationships between the socioeconomic factors and MDD consumption. In addition, the reliance on self-reported information in the DHS survey data might have introduced recall bias or social desirability bias, which is a potential limitation of this study.