Men's nutrition knowledge is important for women's and children's nutrition in Ethiopia

Abstract In an effort to address undernutrition among women and children in rural areas of low‐income countries, nutrition‐sensitive agriculture (NSA) and behaviour change communication (BCC) projects heavily focus on women as an entry point to effect nutritional outcomes. There is limited evidence on the role of men's contribution in improving household diets. In this Agriculture to Nutrition trial (Clinicaltrials.gov identifier: NCT03152227), we explored associations between men's and women's nutritional knowledge on households', children's and women's dietary diversity. At the midline evaluation conducted in July 2017, FAO's nutrition knowledge questionnaire was administered to male and female partners in 1396 households. There was a high degree of agreement (88%) on knowledge about exclusive breastfeeding between parents; however, only 56–66% of the households had agreement when comparing knowledge of dietary sources of vitamin A or iron. Factor analysis of knowledge dimensions resulted in identifying two domains, namely, ‘dietary’ and ‘vitamin’ knowledge. Dietary knowledge had a larger effect on women's and children's dietary diversities than vitamin knowledge. Men's dietary knowledge had strong positive associations with households' dietary diversity scores (0.24, P value = 0.001), children's dietary diversity (0.19, P value = 0.008) and women's dietary diversity (0.18, P value < 0.001). Distance to markets and men's education levels modified the effects of nutrition knowledge on dietary diversity. While previous NSA and BCC interventions predominantly focused on uptake among women, there is a large gap and strong potential for men’s engagement in improving household nutrition. Interventions that expand the role of men in NSA may synergistically improve household nutrition outcomes.

women's dietary diversity (0.18, P value < 0.001). Distance to markets and men's education levels modified the effects of nutrition knowledge on dietary diversity. While previous NSA and BCC interventions predominantly focused on uptake among women, there is a large gap and strong potential for men's engagement in improving household nutrition. Interventions that expand the role of men in NSA may synergistically improve household nutrition outcomes. programmes because nutrition is typically perceived to bea woman's domain. In particular, NSA programmes often focus on improving women's nutrition knowledge and empowerment to improve their decision-making power for food purchases and allocation of nutritious food (Ruel, Alderman, Maternal, & Child Nutrition Study, 2013;Ruel, Quisumbing, & Balagamwala, 2018).
One study found that engaging husbands during pregnancy resulted in higher dietary diversity among women in Bangladesh, but it is unknown whether these effects are sustained after pregnancy or observed among non-pregnant or lactating women (Nguyen et al., 2018). Women's empowerment, however, cannot be achieved without equitable contribution from men, especially in their roles as fathers, husbands, household heads and, more importantly, prominent players in decision-making on income, food purchases, and consumption (Engle, 1997). Despite the central role of men, very few studies have evaluated the impact of men's engagement on household nutrition, including diets and nutritional status of women in low resource settings (Schneider & Masters, 2018). Highlighted in Figure 1 are the hypothesized pathways from nutrition knowledge to household nutrition outcomes based on existing literature (green), current analysis (purple) and proposed future research (grey). We have aligned some of these pathways with theongoing and innovative work on Women's Empowerment in Nutrition dimensions, with a focus on knowledge, agency, and resources (Narayanan, Lentz, Fontana, De, & Kulkarni, 2019).
An innovative study in Northern Ethiopia found that fathers' nutrition knowledge and education was associated with higher dietary diversity among children but did not account for mothers' knowledge or education (Bilal et al., 2016). Taken together, these studies suggest that nutrition knowledge (mostly women's) is necessary but not sufficient for optimal nutrition outcomes (mostly children's) and that there may be other contributing factors such as education (women's and men's), household wealth and access to markets to leverage higher gains from knowledge to nutrition outcomes (Burchi, 2010;Hirvonen, Hoddinott, Minten, & Stifel, 2017;Ruel et al., 1992). Additionally, the importance of the nutrition knowledge of other family members, such as grandparents, for child outcomes has been explored extensively (Karmacharya, Cunningham, Choufani, & Kadiyala, 2017). Informational flow between grandparents and mothers occurs; however, nutrition knowledge flow between mothers and older children (siblings) on their own nutrition or younger children's nutrition outcomes remains to be explored.
Distillation of these studies conducted in low-income settings points to three substantial gaps. First, the impact of men's (fathers'/ spouses') nutrition knowledge on women's and children's nutrition outcomes remains under-studied. Second, an understanding of how men's and women's nutrition knowledge within a household are associated for optimal nutrition outcomes remains unknown. Lastly, components of nutrition knowledge associated with the highest gains in nutrition outcomes need to be identified.
To address these research gaps, we used data from Agriculture to Nutrition (ATONU) study (Clinicaltrials.gov identifier: NCT03152227) --a cluster randomized trial conducted in Ethiopia. The main objectives of this paper are: (1) to describe men's and women's nutrition knowledge and agreement between these two within a household; (2) to examine how nutrition knowledge of both men and women is associated with households', children's and women's dietary diversity after adjusting for men's and women's education, household wealth and size, and village-level clustering; and (3) to identify components of nutrition knowledge with the highest effect size on nutrition outcomes.

Key messages
• There is very little focus on men's role in women's and children's dietary outcomes in low-income settings.
• Within households, men and women have high knowledge and agreement on optimal breastfeeding practices.
However, there is low knowledge and agreement between men and women on complementary feeding, iron-deficiency anaemia and vitamin A deficiency.
• Two components of nutrition knowledge (dietary and vitamin) among men and women were associated with higher dietary diversities of women, children and households.
• Men's nutrition knowledge had significant, positive and additive associations with households', children's and women's dietary diversity after adjusting for household wealth, women's education and nutrition knowledge.
• Targeted research exploring how nutrition knowledge is gendered and how to engage men in nutrition programming may lead to better outcomes.

| Study setting
We used data obtained from ATONU, a cluster randomized trial that was nested within the African Chicken Genetic Gains (ACGG) project and has been described previously (Ambikapathi et al., 2019;Dessie, 2016). The trial began in 2016 with 21 months of intervention activities across four regions of Ethiopia, including Tigray, Amhara, Oromia and Southern Nations, Nationalities, Peoples' Region (SNNPR). Interventions included the introduction of 25 chickens of improved breeds per household (arm 1, 'ACGG'); behaviour change communication on women and children's nutrition, water, sanitation, hygiene, and women's empowerment, plus the 25 improved chickens (arm 2, 'ACGG + ATONU'); and lastly, a no intervention arm (arm 3, 'control'). Villages, the primary sampling units, were randomly selected, and stratified by district and agro-ecological zone.
At the baseline evaluation, 2,117 households were enrolled in the study. Households meeting the following inclusion criteria were eligible to be enrolled in the study: (1) have a woman of reproductive age (18-45 years), (2) provide informed consent, and (3) participated in chicken farming for the last 2 years and currently have less than 50 chicken (same criteria for the ACGG programme). Surveys were administered to the household head and one woman of reproductive age. Among households with children under 36 months, one eligible child was picked at random for anthropometry, morbidity and dietary diversity assessments.
The current analysis uses data from the midline evaluation because nutrition knowledge surveys were only added at this evaluation. The survey was conducted from July to August 2017 on 2,042 households (75 were lost to follow-up from baseline). For the purposes of this analysis, only households with a married couple (e.g., male household heads married to women) who answered the nutrition knowledge surveys were included; hence, 646 households were excluded for the following reasons: 274 woman-headed households, 347 respondents in a non-marital relationship with the household head and 25 surveys with missing data. The excluded 274 women-headed households did not vary significantly with regard to the three main outcomes (women's, children's and household dietary diversity scores). In total, 1,396 households with 743 children were included in the analysis.
Physical access to market in terms of duration (minutes to travel from the household to the market) was available only among 84% of the sample population and was limited to three regions (Amhara, Oromia and SNNPR) at the midline evaluation; therefore, market access was included in a subset analysis. Food security was measured using the Household Food Insecurity Access Scale (HFIAS; Coates, Swindale, & Bilinsky, 2007). WHO/UNICEF definitions (2015) were used to estimate the prevalence of improved access to water and sanitation. Household wealth quintiles were developed based on assets, land ownership, and household characteristics (Ambikapathi et al., 2019).

| Key exposures: Nutrition knowledge definitions
Nutrition knowledge of the study participants was assessed using the Food and Agriculture Organization's (FAO) nutrition-related knowledge, attitudes and practices questionnaire (Marías & Glasauer, 2014).
F I G U R E 1 Hypothesized pathways from nutrition knowledge to nutrition outcomes Out of 13 available modules, we used five modules on breastfeeding, infant feeding, nutrition during pregnancy and lactation, iron deficiency and vitamin A deficiency for analysis.. These questions have multiple correct answers listed. Responses were recorded by the survey team if the respondent gave one of the listed answers; responses not listed were entered as text in the 'other' category and were analysed for correctness. Responses within knowledge questions were summarized. For example, there are six correct answers for 'ways to provide good nutrition for pregnant/lactating women' (eating more food, eating more at each meal, eating more frequently, eating more protein-rich foods, eating iron-rich foods and using iodized salt for preparing meals; Marías & Glasauer, 2014). Each item was given 1 point, yielding a maximum possible score of 6. In total, there were four nutrition knowledge variables per woman and man: (1) ways to provide good nutrition for pregnant/lactating women, (2) ways to improve diets for children, (3) knowledge of vitamin A-rich foods and (4) knowledge of iron-rich foods.
Because these four knowledge variables were highly correlated with each other, exploratory factor analysis was utilized to distil nutrition knowledge variables ( Figure 2d). Previous research assessing mothers' knowledge of child nutrition have used similar data reduction approaches (Fadare et al., 2019;Hirvonen et al., 2017). Based on iterative factor analyses (run separately for women and men), two factor models were used, and they explained approximately 75% of the variance in the distilled nutrition knowledge variables. Factor loadings and scores are presented in Table S1. Exploratory factor analysis on nutrition knowledge variables uniquely loaded on two sets of factor groups (factor loadings > 0.3) that were similar for both men and women. This included (1) a 'dietary knowledge' factor, which had high factor loadings on procedural knowledge to improve nutrition for women and children and (2) a 'vitamin knowledge' factor, which had high factor loadings on food groups that are rich with vitamin A or iron (Velardo, 2015). Standardized regression scores for men and women were used as the main nutrition knowledge exposures.

| Key outcome variables
There were three main outcome variables: household dietary diversity scores among households (HDDS, 1-month recall, 10 food groups), children's dietary diversity (CDDS, 1-day recall, seven food groups) and women's dietary diversity (MDD-W, 1-day recall, 10 food groups; FAO & FHI 360, 2016;World Health Organization, 2010;Swindale & Bilinsky, 2006). Less than 5% of the sampled women mentioned that day of dietary data collection was a holiday, whereas 24% mentioned they fasted (did not consume animal source foods according to the Ethiopian Orthodox tradition). There were no significant differences in MDD-W by fasting, likely because of very low intakes of animal source foods. We made a change to HDDS by extending the recall from one day to one month to examine typical food access and because there was low food diversity in these settings. Finally, to examine the specificity of knowledge of food groups to a behaviour, we evaluated the impact of knowledge factors on consumption of individual food groups for women.

| Statistical analysis
For comparison of intervention arms, joint F tests were obtained from generalized linear mixed models adjusting for clustering at the village level. Linear polynomial regression was used to visualize the relationships between dietary diversity scores and knowledge variables ( Figure 2). Spearman rank correlation was used to examine correlations within the eight nutrition knowledge variables. Mixed effects linear and logistic regression models adjusting for village-(kebele, lowest administration unit in Ethiopia) level clustering were used to evaluate the associations between exposures and continuous and binary outcomes. All models were adjusted for household size, wealth quintiles, woman's age and education, man's age and education and the four geographical regions. Models with CDDS were adjusted for child age. Education is often associated with nutrition literacy and uptake1992, and therefore analysis examining the interaction between education and nutrition knowledge was explored in the multivariable models 1992. Treatment arms were not significant in all models, therefore removed from the main models. Summary data are presented below as median with first and third quartiles (interquartile range [IQR]: Q1, Q3) or as percentages.

| Ethical considerations
The study protocol was approved by the Institutional Review Board

| RESULTS
The median age of women included in this analysis was 34 years (IQR: 28, 39), and over half (60%) of women had no schooling, whereas the median age of men was 40 years (IQR: 35, 48), and a quarter (27%) of the men had no schooling (Table 1). Median age of the children was 22 months. Women in the control arm were on average younger, by 3 years, than women in the intervention arms. Seventy-nine percent of the households had access to improved water, whereas only one third of households had access to improved sanitation. The median time to the closest market was 45 min (IQR: 30, 60) and about half of the households reported that they attend the markets weekly. Half (52%) of the households reported having food access security.
Median household dietary diversity scores were four food groups in ACGG and control arms, while the ACGG + ATONU arm had five food groups. The top five food groups consumed by the households in the last 30 days were grains (94%), legumes (69%), oils and fats (57%), dairy (42%) and eggs (40%). Less than 10% of women met the recommended dietary diversity (at least five food groups out of 10). Consumption of individual food groups for women are summarized in Table 1. Besides staples, women most commonly consumed legumes and green leafy vegetables, while very few women reported consuming meat, nuts or other vitamin A-rich produce (mostly vitamin A-rich vegetables) in the previous 24 h. Besides staples, children consumed foods from the fruits and vegetables food groups, followed by vitamin A-rich foods, and other fruits and vegetables. Both women and children rarely consumed meat. Less than 7% of women and 11% of children had consumed eggs in the previous 24 h. Neither dietary diversity nor the consumption of individual foods was significantly different across treatment arms at midline evaluation for women and children. There were regional differences in diets among women, children and households (see Table S2). Median HDDS and CDDS were five and three food groups in Amhara and Oromia. While in SNNPR and Tigray HDDS and CDDS were lower by one food group for HDDS and CDDS.. We saw similar trends in MDD-W with SNNPR having one less food group compared to Tigray, Amhara and Oromia regions.  Table 2 summarizes the nutrition knowledge responses between men and women within a household. Agreement within a household illustrates the knowledge gaps among men and women from the same households. In general, over 80-90% of men and women have high knowledge on exclusive breastfeeding and optimal breastfeeding practices. However, knowledge on food groups and dietary practices to improve nutrition among children and women is very low. There is also higher discordance of knowledge within households on nutrition practices related to women and children and on knowledge of foods rich in T A B L E 1 Demographics and main variables of interest from the ATONU study midline evaluation, July to August 2017, Ethiopia Summary data are either presented as median with quartiles 1 and 3 (Q1, Q3) or percentages within treatment arms with sample size in parentheses. b "Primary 1" refers to 1-5 years of schooling; "Primary 2" refers to 6-9 years of schooling; "Secondary 1" and "Secondary 2" refer to 10-17 years of schooling. specific nutrients. For example, more than 45% of men and women have heard of vitamin A deficiency, but in only 27% of households both individuals have heard of vitamin A deficiency.  Table 4). Men's dietary knowledge was independently associated with HDDS, even after adjusting for women's dietary knowledge and education. Age of both parents and household size was not associated with HDDS.

| Does education modify the effect of nutrition knowledge on dietary diversity scores?
Interaction between nutrition knowledge and education varied by outcome and gender. For MDD-W, there was no significant interaction observed between women's education and their dietary knowledge. However, significant interaction effects were observed for men's education and nutrition knowledge on MDD-W. Among men who attended a religious school or adult literacy programmes, rather than typical formal education, higher nutrition knowledge was associated with significantly lower MDD-W scores among women (see Figure S1). These households represent 10% of the sample population. In these same households, child dietary diversity scores were also lower by 0.  Figure S2. Here, increasing knowledge among fathers was significantly associated with higher dietary diversity among children, but only among households where women had lower standardized dietary knowledge scores (factor scores below 0), which represented 50% of sample population.

| How does nutrition knowledge affect consumption of food groups?
Overall, men's dietary knowledge was associated with significantly higher odds of women consuming dairy, vitamin A-rich foods and dark green leafy vegetables, and the odds ratio varied for different food groups; that is, the effect of knowledge on consumption differed by food group (see Figure 3). Similar trends were observed for women's dietary knowledge. Vitamin knowledge among both men and women was associated with increased odds of women consuming vitamin A rich produce and dark green leafy vegetables.

| DISCUSSION
The diets of women and children (and households generally) were very poor in this rural population in the four most populous regions of Ethiopia; only 9.4% of women and 26.7% of children met the minimum recommendation for dietary diversity. Consumption of animal source foods was low for both women and children. Knowledge of breastfeeding practices was above 80% among both men and women, possibly due to the extensive programming of Alive and Thrive in these same four regions (Menon, Rawat, & Ruel, 2013) and availability of the national health extension programme.
However, knowledge on dietary practices to improve vitamin A or iron intake remained poor, with higher discordance in knowledge between men and women of the same household. Overall, men's and women's nutrition knowledge had a positive relationship with the household's dietary outcomes.
T A B L E 2 Summary of nutrition knowledge questions and correct answers from women and men, agreement within household, and factor analysis grouping   Table S2) Significant (see Figure 3 and Table S2  has a positive and significant association. This may be because men have higher education compared with women, which may result in higher vitamin knowledge than women. We also noticed that knowledge variables of the father and mother appeared to attenuate the effect size of each otheron child dietary diversity score, possibly due to high correlation between men's and women's knowledge (r = 0.5; see Figure 2).
Education, wealth, and access to markets are common mediators and modifiers of women's nutrition knowledge on child nutrition outcomes (Burchi, 2010;Hirvonen et al., 2017;Onyeneke et al., 2019;Ruel et al., 1992). In this analysis, there were no interaction effects between education and knowledge for either parent on children's dietary diversity.
There are several explanations for the observed results. First, most of the sampled population had a low education level; for example, 60% of mothers in this analysis had no schooling, and an additional 20% had fewer than 5 years of schooling. These results are similar to other studies (Bilal et al., 2016;Hirvonen et al., 2017;Oduor et al., 2018) where the majority of caregivers had low education. Second, substitution (and collinearity) between parents in the same household for knowledge and education attenuated the effect size of these factors individually. In samples where there is heterogeneity in education levels, women's education appears to have a larger impact than men's education on dietary diversity (Onyeneke et al., 2019;Ruel et al., 1992) and other nutrition outcomes (Alderman & Headey, 2017). A previous study in Ethiopia found that fathers' education appears to have a small positive effect (0.09 food groups) on the child's dietary diversity score (Hirvonen et al., 2017). In this study we see similar results, where fathers' education levels were not associated with CDDS, except among fathers who had religious schooling or had attended adult literacy programmes, in which case these households had lower CDDS. The percentage of men who went to religious school or literacy programmes is less than 10% (n = 115). In these households, men are at least 8 years older than the rest of sample population, but no other differences in demographics were observed. In these households, consumption of vitamin A rich produce is generally lower for both women and children. We also note that nutrition knowledge between men and women does seem to attenuate each other's effectsize, when both are added to the model (C-model 5), perhaps due to the positive correlation between these variables.
In context. In this analysis, we make the assumption that the measurement error with this instrument was similar between genders, regions, and education levels.
This study is novel in that it considers men's education, age and nutritional knowledge along with women's education, age and nutrition knowledge, to examine effects on women's and children's dietary F I G U R E 3 Results from mixed effects logistic regression of consuming individual food groups among women. All models adjusted for household size, household wealth quintile, women's woman's age, man's age, woman's education, man's education, region and kebele-level clustering (treatment effect was not significant). DGV: dark green vegetables; Vitamin A: vitamin A rich produce (including both vegetables and fruits that are rich sources of vitamin A) outcomes, assessing and specifically estimating the additive effects of men's characteristics for household nutrition outcomes. We also focused on specificity of exposures, such as the impact of knowledge of dietary practices on specific dietary behaviors , rather than longerterm effects on nutritional status. Finally, we show results from multiple models to evaluate the change in coefficients of key exposures on outcomes. Although not causal, these results are useful for testing and generating new hypotheses on pathways (grey arrows in  (7) local conceptualization of nutrition knowledge and practices and differences in these frameworks by gender, age (adolescents, school-aged children and grandparents), and stakeholder type(food vendors, health care workers, community health workers and leaders). In future analyses, we aim to address the first three questions longitudinally, incorporating findings from a qualitative study that interviewed men and women about men's engagement in nutrition and caregiving. We invite other researchers to focus on these identified topics, especially using existing datasets from NSA programmes, to pursue the imperative and achievable target of optimal women's and children's nutrition outcomes through men's engagement..