Women's empowerment, children's nutritional status, and the mediating role of household headship structure: Evidence from sub‐Saharan Africa

Abstract We aimed to examine the association between women's empowerment and childhood nutritional status while accounting for the mediating role of household headship structure. Cross‐country, cross‐sectional quantitative data from the most recent Demographic and Health Surveys (2015–2018) were used. Women's empowerment was measured as a composite index of participation in household decision‐making, attitude towards domestic violence, and asset ownership. Childhood nutrition status was measure as anaemia (haemoglobin concentration < 110g/L), stunting (height‐for‐age z‐scorescore <−2) and the co‐occurrence of anaemia and stunting. Applying the Lewbel two‐stage least squares, women's migration status was used as an instrumental variable. We used data on 25,665 woman‐child dyads from eight sub‐Saharan African countries: Burundi (2016), Ethiopia (2016), Guinea (2018), Malawi (2016), Mali (2018), Zimbabwe (2015), Uganda (2016), and Tanzania (2015). The women were in their reproductive ages (15–49 years) and children were under 5 years old. The findings showed that an increase in women's empowerment index reduces children's likelihood of being anaemic and having a co‐occurrence of anaemia and stunting [coeff (SE), −0.114 (0.025) and −0.072 (0.032), respectively]. Specifically, an increase in asset ownership or decision‐making dimensions of empowerment significantly reduces the likelihood of anaemia and the co‐occurrence of anaemia and stunting among children. Children of empowered women from male‐headed households were more likely to be anaemic and be concurrently anaemic and stunted compared to their counterparts whose mothers were from female‐headed households. Interventions designed to improve childhood nutrition through women's empowerment approaches need to consider asset ownership and instrumental agency of women while acknowledging the mediating effect of household headship typology.


| INTRODUCTION
Women's empowerment (autonomy) is considered the expansion of women's ability to make strategic life choices (Kabeer, 1999). Interest in women's autonomy and empowerment has gained increased attention over the past three decades, not only for its potential effect on women themselves but also on their families, communities, and national development (Zuccala & Horton, 2018). Scholarship, particularly in sub-Saharan Africa (SSA) links women's empowerment to household and individual outcomes such as health and nutrition (Atiglo et al., 2020;Atiglo & Codjoe, 2019;Onah, 2020;Quisumbing et al., 2021). This is understandable, given that women are the primary caregivers and major gatekeepers in their households when it comes to food and nutrition in SSA (Hatch & Posel, 2018;Mkhwanazi et al., 2018). Findings on the relationship between women's empowerment and child nutrition outcomes are however inconsistent, thus, the need for examination of some nuances in this relationship (Cunningham et al., 2015;Onah, 2020;Santoso et al., 2019).

| Women's empowerment and children's nutrition outcomes
Generally, gendered social and cultural norms influence intrahousehold relationships and are also known to influence child health and nutrition outcomes (Mwaseba & Kaarhus, 2015;Seebens, 2010). Central to research on the relationship between women's empowerment and households' health and nutrition outcomes is the varying interest and expenditure when resources are in the hands of women or men. The inquiry is usually to determine the importance of women's control of resources to the health and well-being of children who are their primary responsibility when it comes to caregiving. Current scholarship suggests that women with more say or control over resources are more likely to improve the nutrition of their children compared with their counterparts who are less empowered and defer to the will of their husbands or significant others in the household (Gribble & Preston, 1993). Generally, women are more likely than their male counterparts to use household cash to buy food and provide health care for their children (Porter, 2016;UNICEF, 2011), which has a positive impact on household-level calorie availability and health outcomes. The control of resources by women may reflect positively on household nutrition and well-being (Santoso et al., 2019), however, women are generally disproportionately found to lack some assets and wealth which may deprive them of the benefits associated with their control of such. Conversely, previous studies have shown household structure, such as headship structure to influence children's nutritional status (Mikalitsa, 2015). However, these studies In answering the above questions, this study contributes to current scholarship in the following ways. First, we add to the literature on the determinants of childhood malnutrition, particularly the double burden of undernutrition, which is specifically the cooccurrence of stunting and anaemia. This is because previous investigations on the effect of women's empowerment on children have been limited mostly to the relationship between women's empowerment and single malnutrition conditions (Mikalitsa, 2015).
Given the similarity of contributing factors of malnutrition conditions and the extra burden of double malnutrition conditions, current literature argues for the need to examine factors contributing to the presence of double malnutrition conditions in individuals Christian et al., 2023;Christian & Dake, 2022

Key messages
• The mediating role of household headship in women's empowerment and childhood nutrition nexus was explored in eight sub-Saharan African countries.
• Overall, an increase in women's empowerment significantly reduces anaemia and the co-occurrence of anaemia and stunting in children.
• Specifically, women's autonomy in asset ownership and decision-making reduces anaemia and the co-occurrence of anaemia and stunting in children.
• Children of empowered women belonging to maleheaded households had poorer nutritional status than those in households with empowered female heads. (Cunningham et al., 2015). This study examines the effect of different dimensions of women's empowerment on childhood nutrition measured as stunting, anaemia and the co-occurrence of stunting and anaemia.  (2015). These were countries that collected data on all variables of interest (migrant status, household headship, women's empowerment, and children's nutritional outcomes). In each country, the prevalence of the double burden of malnutrition is over 30% (33% in Zimbabwe to 58% in Burundi). The DHS is a cross-sectional, nationally representative health survey conducted in selected lowand middle-income countries approximately every 5 years. Respondent sampling, selection, and administered questionnaires are standardised to make DHS data comparable over time and across countries. The DHS employs a stratified cluster sampling procedure to select census enumeration areas based on a probability proportional to the size of the enumeration area. Households are then randomly sampled within each of the selected enumeration areas.

| Household headship
A household is considered a unit or group of individuals living together and sharing similar food and other essential living arrangements (Randall et al., 2015). Households were classified as either male-headed or female-headed households. This was based on respondents' answers to the question: 'Please tell me the name of each person who usually lives here, starting with the head of the household'. The household head is considered an adult male or female who is responsible for the organisation and care of members of the household and is considered the head by members of that household.

| Women's empowerment
A broad body of scholarship demonstrates that women's empowerment is linked to their role or ability to partake in important household decisions, their likelihood of suffering from or tolerance of domestic violence and whether they own assets in the family they F I G U R E 1 A conceptual framework for the intrarelationships between women's empowerment, household headship and children's nutritional outcomes. CHRISTIAN ET AL. | 3 of 13 belong to. In this study we used three main domains of women's empowerment adopted by the standard DHS questionnaire: 1. Women's role in household decision-making (based on responses to questions relating to women's participation or say in six different types of household decisions on (a) woman's own earning (b) what to do with the money the husband earns, (c) visits to family or relatives, (d) large/major household purchases, (e) daily needs major purchases, and (f) respondent's health care). A composite trichotomous measure was obtained by summing women's responses into low (score of 0), moderate (score of 1-3), and high (score of 4-6) levels of participation in decision-making.
2. Women's attitude towards domestic violence, as represented by questions asking women whether they agree or disagree with five scenarios in which a husband is justified in beating their wife (when she burns food; argues with her husband; goes out without telling her husband; neglects children; and refuses husband sex). A woman that says 'NO' is considered empowered in that domain and given a score of '1'. A woman that says 'YES' is considered not empowered in that domain and given a score of '0'. Their 'no' or 'yes' responses were summed to form a three-category variable consisting of low (0), moderate (1-3) and high (4 and 5) autonomy categories.
3. Asset ownership by women, measured by asking women whether they owned a house alone or jointly; owned a land alone or jointly.
Their responses were summed to create a variable which was categorised into no (0 positive response), moderate (1 positive response) and complete ownership of assets (2 positive responses).
Categorising the various empowerment indicators into low, moderate, and high has been adopted by several researchers that seek to examine how women's empowerment influences household outcomes (Asim et al., 2022;Rizkianti et al., 2020).
The three individual empowerment domains were utilised in some models in this analysis to assess their differential individual influence on children's nutritional status. A composite variable of women's empowerment was also computed by summing the response categories for the three empowerment domains (participation in household decision-making, attitude towards domestic violence and asset ownership) to obtain a continuous score (0-12) which was used to assess the influence of the levels of empowerment on children's nutritional status.
Higher scores represented higher levels of empowerment. The data included only women who were married or in consensual unions as the questions on women's empowerment in the DHS were only asked to these. Unmarried women were not asked questions on empowerment.

| Children's nutritional status
The DHS collected anthropometric data. Children were classified as stunted if height for age z-scores was less than −2.0 (more than 2 standard deviations below the reference median) (de Onis & Branca, 2016). Stunted children are at higher risk for cognitive impairments, reduced academic achievement, and an elevated likelihood of chronic diseases in adulthood (WHO, 2020). With a drop of capillary blood, a HemoCue (Hemocue Inc.) was used to determine the children's haemoglobin concentration. Children with haemoglobin concentration <110 g/L were considered anaemic. Childhood anaemia is caused predominantly by iron deficiency and is linked to numerous negative consequences including fatigue, weakness, and cognitive, impairments, decreased work capacity, increased susceptibility to infections, physical and cognitive developmental delays. The co-occurrence of anaemia and stunting (CAS) variable was determined by identifying children who had any severity of anaemia concurrently with being classified as stunted.

| Other covariates
Based on existing theory, the covariates selected include (1) child characteristics such as age and sex, (2) mother's characteristics including educational status and age, and (3) household characteristics comprising the total number of household members (household size), access to an improved water source and/or an improved sanitation source and wealth. The categorisation of household water and sanitation sources as improved or unimproved is based on WHO/ UNICEF classification. Households were classified as using unimproved drinking water sources if their primary source of water came from a river, stream, pond, unprotected spring or well, lake, canal, dam, or irrigation channel and they did not treat the water (such as boiling, using bleach or chlorine, water filtering, solar disinfection, or letting the water stand and settle). Having an improved water source was defined as having access to any one of the following: piped water, public taps, standpipes, tube wells, boreholes, protected dug wells and springs, and rainwater (WHO/UNICEF, 2015). Households with improved sanitation are those with access to a pour-flush toilet, ventilated improved latrine or a composting toilet or a pit latrine with slab, and those without any of the improved facilities are considered to be using an unimproved facility. Household wealth was generated using assigned asset weights that were generated from a principal components analysis to create standardised asset scores, which were then categorised into terciles (Rutstein et al., 2004).

| Econometric framework and strategy
We employed the probit regression model to examine the relationship between women's empowerment and child nutrition, where 1 represents a child with a malnutrition condition (i.e., an anaemic child/a stunted child/a child who is both anaemic and stunted) and 0 otherwise, the generalised form of the model is presented below: where, Nut ij is the nutritional status of child i in household j; WEmpower kj represents the empowerment index for woman k in household j; X kj , Ψ ij , and η j denotes a vector of characteristics for woman k, child i and household j, respectively; ε ij signifies the randomly distributed error term; β′s represent the coefficients of the regressors and α denotes the constant term.
Estimates from Equation (1) are likely to suffer from omitted variable bias, which tends to render our estimate endogenous. Thus, potential unobservable covariates are likely to correlate with women's empowerment and child nutrition, which can bias our estimates and alter the generalisation of the results. We argue that women's empowerment will be influenced by women's migration status which will affect their children's nutritional outcomes. Though it is expected that current migration flows will lead to more women remaining in rural or urban communities to take over typical male roles, there has been an increase in female migration over the past two decades (Pickbourn, 2018). Migration provides women with the opportunity to build human, social and capital assets. While literature shows that  (Lewbel, 2012). The method is briefly discussed below: where Y 1 represent the outcome variable (child nutrition); Y 2 denotes the endogenous variable, women's empowerment; X′ is the vector of explanatory variables while μ 1 and μ 2 signify the error terms. This technique utilises an identification strategy that relies on information contained in the heteroscedasticity of μ 2 in solving an endogeneity problem in the absence of external instruments. This method assumes The critical assumption under this technique is that there should be no correlation between regressors and heteroscedastic errors. As a further robustness check in addressing endogeneity in nonexperimental datasets, we used the propensity score matching (PSM). In our case, we estimated the average treatment effect (empowered or not) on child nutrition. The average treatment effect is evaluated as follows: where τ represents the average effect of the treatment, θ denotes our dependent variable while ϑ signifies the dummy variable, which equals 1 if the woman is empowered and 0 otherwise. The vector of pretreatment variables is represented by the control variables and denoted by ϖ. Following this formulation, the propensity score, p ϖ ( ) is defined as the likelihood for a child with a malnutrition condition (i.e., an anaemic child/a stunted child/a child who is both anaemic and stunted), given the covariates.  et al., 2012) and therefore, we adopted Baron and Kenny's (1986) stepwise regression method to verify the mediation effect (Baron & Kenny, 1986) The model consists of the following three steps: where M i represents the mediating variable. For M i to serve as a channel of influence, it first needs to correlate with women's empowerment as shown in Equation (6) and render empowerment statistically insignificant or reduce its magnitude when included as an additional variable in Equation (7). The reduction in magnitude can be observed when the coefficient of women's empowerment in Equation (7)

| Ethics of human subject participation
The current study analysed secondary data from the Demographic and Health Survey (DHS). In accordance with the ethics of conducting research with human subjects, voluntary consent to participate in the survey was sought from respondents before being interviewed by fieldworkers who were trained to conduct interviews.
IPUMS DHS granted access to the data.  (Table 1). Table 2 presents the estimates of the relationship between women's empowerment and children's nutritional outcomes (anaemia, stunting, and CAS) in the eight study countries. We adjusted for various household, women, and child characteristics and used a rural-urban dummy variable and country-fixed effects to capture location heterogeneities. In the pooled analysis, women's empowerment was associated with childhood anaemia and CAS but not stunting.

| Women's empowerment and children's nutritional outcome
The estimated coefficients reveal that the likelihood of a child being anaemic was reduced by 1.4% and 0.4% for the CAS with a one-unit increase in the women's empowerment score.

| Robustness check
Making inferences with these estimates may be problematic as they are likely to suffer from omitted variable bias. Omitted variable bias occurs when the right or appropriate controls are not included or missed in a regression model. Migration was used as an instrument to solve the endogeneity between women's empowerment and child nutrition. The empirical results based on the Lewbel 2SLS regression with both internal and external instruments are presented in Table 3.

The Cragg-Donald weak identification Wald test and the
Kleibergen-Paaprk Wald F statistic suggest that our proposed instrument is valid. In effect, migrant women tended to be more empowered compared to their nonmigrant counterparts and women's empowerment reduced anaemia and CAS among children.
These reported results buttress the effect of empowerment on anaemia presented in Table 1.
As a further confirmatory analysis, we employed the PSM (

| Heterogenous analysis
We tested the effect of the various dimensions of women's empowerment (asset ownership, decision-making, and domestic violence) on two of the outcomes (anaemia and CAS; Table 5). An increase in asset ownership and the decision-making dimension of women's empowerment significantly leads to children being less likely to be anaemic (Panel A, Table 5). The effect can be observed to be relatively higher for women with high autonomy.
Children whose mothers had moderate and complete ownership of their assets were 11.5% and 14.4% less likely to be anaemic, respectively. A similar finding was observed for CAS. Children whose caregivers/mothers owned moderate levels of assets were about 7.4% less likely to have a double burden of anaemia and stunting, compared to those whose mothers had no assets. Similarly, children whose caregivers/mothers had high decision-making autonomy were 38.2% and 12.8% less likely to suffer from anaemia and CAS.
However, the domestic violence dimension of empowerment had no impact on either outcome.

| Channel of influence (empowermenthousehold headship-children's nutrition outcome)
The methods applied in the above analysis accounted for the direct impact of women's empowerment on childhood nutritional out- T A B L E 2 A probit model on the association between women's empowerment and children's nutritional outcomes.  Note: Empowerment represents the sum of response categories for the three empowerment domains (participation in household decision-making, attitude towards domestic violence and assets ownership) to obtain a continuous score (0-12). Wealth tertiles were calculated from an assetbased wealth index using assigned asset weights from a principal components analysis to create standardised asset scores. Diarrhoea was defined as 'three or more loose or watery bowel motions in a 24-h period, as reported by the child's mother or caregiver at any time during the 2 weeks before the interview' (WHO, 2013).
T A B L E 3 A Lewbel 2SLS for the effect of women empowerment on child nutritional status. Note: Standard errors in parentheses.
children's nutritional status through other channels. Table 6 presents an analysis of a possible channel through which women's empowerment could influence child nutrition using Baron & Kenny's (1986) stepwise approach. Women's empowerment was associated with the household structure. The results showed that the household headship structure was significantly associated with anaemia and CAS in children, respectively.
However, the inclusion of household headship structure as an additional control variable in these models reduced the magni-  (Sony et al., 2020).
The results show that migration influences children's nutritional outcomes (anaemia and CAS) via women's empowerment. de Brauw et al. (2021) confirmed that women's empowerment was more likely to increase when female household members migrated than when male members of the same household migrated (de Brauw et al., 2021). Despite the possibility of increased their vulnerability to exploitation and abuse, extant research shows that the effect of migration on women is that they are more empowered, although not in all domains.
Migration can be liberating for women and an opportunity for them to escape social restrictions or gender discrimination (Deshingkar & Grimm, 2005;Hugo, 2000). Women can have better access to a variety of education and work options because of migration, which can contribute to reducing income and wealth inequality directly or indirectly (Eryar et al., 2018).

| CONCLUSION AND RECOMMENDATION
This paper investigated the effect of women's empowerment on children's nutritional outcomes, mediated by household headship and women's migration status. Findings reinforce the importance of maternal autonomy and empowerment when targeting interventions to improve the nutrition of children in households. Attention however must be paid to the domains of empowerment that intervention seeks to improve and the household dynamics, particularly the headship structure in which women and children find themselves. Programme managers will have to design interventions to promote women to participate in decision-making processes, especially those related to health care and nutrition, which can help them make informed decisions that benefit their families. Additionally, there is a need to implement policies that promote gender equality. Given the robustness of our results, we could say that investing in women's empowerment may not only be essential for the well-being of women but also to ensure better nutrition for their children.