Impact of agricultural interventions on food and nutrition security in Ethiopia: Uncovering pathways linking agriculture to improved nutrition

Abstract Achieving sustainable food and nutrition security calls for multi-sectoral coordination mechanisms in implementing, among others, nutrition-sensitive agricultural interventions. Despite progresses made in evaluating the effectiveness/impact of these interventions in some countries, very little is known about the nature and causal effect of these interventions in Ethiopia. Therefore, this systematic review aims to provide evidence on the impact of nutrition-sensitive agricultural interventions and pathways linking agriculture to improved nutrition. Based on a comprehensive advanced literature search using keywords in various databases and setting inclusion/screening criteria, we identify 25 relevant studies conducted between 2008 and 2018. Overall, we find positive effects on knowledge, attitude, and practice of complementary feeding and economic indicators (income, poverty and asset accumulation). We also find an overwhelming positive effect on food consumption, dietary diversity, and food security indicators. However, we also note mixed effectiveness in relation to child anthropometry. The most important pathways linking agriculture to improved nutrition center on its role as a source of food and income, and its effect on food prices/market. Hence, enhancing community-based nutrition education, accelerating adoption of improved technologies, improving access to farm inputs and technical information, integrating farmers to markets, and boosting enabling environments and capacities should be emboldened.


PUBLIC INTEREST STATEMENT
Agriculture continues to be the life support mechanism for millions of smallholders across the globe. In addition to enabling food security, the sector is expected to positively impact nutritional security. Hence, there is a lot of attention given to uncover the link between agricultural investments and changes on food security status, household dietary diversity, and overall economic development. In developing countries like Ethiopia, however, there is a scanty systematic and empirical literature available to shed light on this role of agriculture. This contribution is, therefore, an attempt made to illuminate the role nutrition-sensitive agricultural interventions play in the context of Ethiopian livelihoods.

Introduction
There have been growing global consensus that multi-sectoral approaches and coordination mechanisms should be followed to address the debilitating impacts of malnutrition (FAO, 2017;Sassi, 2017;World Bank, 2015). In particular, nutrition-sensitive interventions in the food and agricultural sector are believed to play a pivotal role (Beyero, Judith, & Amanda, 2015;FAO, 2017;Ruel & Harold, 2013) alongside the nutrition-specific interventions. According to FAO (2017), nutrition-sensitive agriculture entails the sustainable production of diverse, economically affordable, culturally acceptable, and safe animal-/plant-based foods in adequate quantity and quality in order to satisfy the dietary needs of populations. Such interventions can be a game-changer contingent upon favourable/enabling environments (Beyero et al., 2015;Ruel & Harold, 2013), appropriate policy and governance structures (Turner et al., 2013), and enhanced local implementation capacities (Warren and Edward, 2017).
In this regard, there have been progresses made in evaluating the effectiveness/impact of such interventions (see, for instance, Rahman & Islam, 2014;Yosef, Andrew, Barnali, & Stuart, 2015;Fiorella, Rona, Erin, & Lia, 2016). These studies provide evidence on how agriculture can potentially influence nutrition, entry points into nutrition-sensitive agriculture, specific interventions with a greater effect on nutritional outcomes, and challenges in targeting vulnerable segments of the society. Nevertheless, gaps still remain to draw definitive conclusions. For example, there are studies that show weak or little effect on nutritional outcomes (Haddad, 2013;McDermott, Myriam, Julien, & Nancy, 2013;Ruel & Harold, 2013;Webb & Eileen, 2014) or mixed efficacy of such interventions in some developing countries (Dorward, 2013).
A recent analysis shows weaknesses in study design, survey tools and measurements as the major contributing factors (Webb & Eileen, 2014). Realizing these gaps and variations in the design and implementation of nutrition-sensitive agricultural interventions across countries and regions, there is a need to generate additional knowledge to build a strong evidence base. In particular, methodologically rigorous and high-quality evaluations are required (Beyero et al., 2015;Girard, Julie, Corey, & Olafunke, 2012;Ruel & Harold, 2013). Furthermore, location-specific research is needed on many of the agriculture-nutrition pathways and processes (Yosef et al., 2015) and diverse outcome indicators . Therefore, the present study is one that addresses these gaps in order to provide specific and policy-relevant recommendations. In this regard, Ethiopia provides a special case because despite the high commitment from the government and other stakeholders-reflected through the National Nutrition Programme (NNP), increased budget, and promotion of nutrition-sensitive interventionsundernutrition remains to be a serious challenge (see Section 2 for details). In addition, there is little recent empirical evidence that evaluates the impact pathways especially after the introduction of the NNP in 2008. Hence, this study is aimed at identifying nutrition-sensitive agricultural interventions in the country, evaluating their impact primarily on food and nutrition security, and identifying key pathways linking agriculture to enhanced food and nutrition outcomes.

A brief review of agriculture-nutrition linkages in Ethiopia: progresses and gaps 1
According to the Global Nutrition Report (2017), worldwide 23% of children under five years of age were stunted and 8% were wasted in 2016. The data for Eastern Africa imply worsened conditions especially for stunting (37%). In Ethiopia, both stunting (38%) and wasting (10%) were higher. Huge gaps also remain in adolescent and adult nutritional status. Several interventions have been initiated to address the underlying and immediate causes, and thereby curb the effect, of undernutrition in the country. The National Nutrition Strategy (NNS) was launched in 2008 and operationalized through the National Nutrition Programme (NNP). The NNP aims at mainstreaming nutrition into the programming of various sectors. The NNP has five strategic objectives (Federal Ministry of Health (FMOH) Government of Ethiopia, 2013): (1) improve nutritional status of women  and adolescents (10)(11)(12)(13)(14)(15)(16)(17)(18)(19), (2) improve nutritional status of infants, young children and children under-five, (3) improve delivery of nutrition services for communicable and noncommunicable/lifestyle-related diseases, (4) strengthen implementation of nutrition-sensitive interventions across sectors, and (5) improve multisectoral coordination and capacity. In the 2015-2020 revised NNP-II, the government is committed to reduce stunting to 26% by 2020 and to allocate additional domestic funding of US$15 million (FMOH, 2013).
The Ministry of Agriculture and Natural Resources (MOANR) also included nutrition-sensitive provisions in the Productive Safety Net Program (PSNP) and Agricultural Growth Program II (AGP-II). In addition to ensuring food security and better livelihoods, the current phase (2015-2020, i.e., PSNP4) focuses on addressing undernutrition through multisectoral programming (World Bank, 2015). The MOANR also drafted a strategic plan for nutrition-sensitive agriculture in October 2016 (The FDRE MONR, 2016). This strategy aims at: (1) increasing production of diverse nutrient-dense foods, (2) increasing dietary diversity, (3) developing agricultural markets, and (4)  Despite these progresses, there are still widespread challenges associated with malnutrition in the country. The country has the seventh highest number of stunted children in the world (European Commission, 2017), and according to the Global Burden of Disease Study, childhood stunting, wasting and underweight are the major reasons for deaths of children under five years of age due to diarrhoea and other common infections (Deribew et al., 2010). Moreover, according to Save the Children (2014), the cost of malnutrition to the national economy in 2013 was US$ 4.7 billion (i.e.,16.5% of GDP). So, what is the contribution of nutrition-sensitive agricultural interventions in the fight against malnutrition? This is one of the central questions that we investigate in this review.

Conceptual framework
A careful consideration is made in the development of a conceptual framework that illustrates the linkages between agriculture and nutrition and serves as a guide in the review process. In this regard, the conceptual frameworks in previous reviews, such as in Masset, Lawrence, Alex, andJairo (2011), Gillespie, Harris, andKadiyala (2012), Hawkes, Turner, and Waage (2012), Ruel and Harold (2013), and Kadiyala, Jody, Derek, Sivan, and Stuart (2014) are critically evaluated. In addition, conceptual frameworks in studies, such as Girard et al. (2012), Turner et al. (2013), Webb (2013), and  are assessed. Consequently, the framework by Masset et al. (2011), with the addition of seven impact pathways (Webb, 2013), is adapted in our investigation.

Comprehensive literature search
As a second stage in the review process, past and current publications are searched to locate and list all the relevant interventions, pathways and processes. A similar approach was followed in previous studies (e.g., Girard et al., 2012;Masset et al., 2011;Turner et al., 2013;Webb & Eileen, 2014). The search is done in February 2018 on the following databases/search engines: PubMed, Web of Science (All Databases), EconLit, Popline, Scopus, The Lancet, Proquest, AGRIS, ELDIS, Ideas/repec, World Bank (Jolis), DFID, IFPRI, FAO (fapda) and Food Security Portal. The keywords used in the search relate to: agriculture, nutrition, food production, rural development, homestead garden, livestock, dairy, poultry, crop, food security, anthropometry, child growth, diet, health, and sustainable development.
The studies obtained through the search procedure are scrutinized for relevance through setting inclusion criteria. Such criteria were set and employed by previous studies, such as by Masset et al. (2011), Girard et al. (2012, Turner et al. (2013), and Kadiyala et al. (2014). The present study considers the following inclusion criteria: (1) time period: 2008-2018 (the year 2008 was chosen in order to coincide with the onset of the Ethiopian National Nutrition Programme); (2) studies that exclusively state improvements in nutritional outcomes as an objective; (3) studies that evaluate the impact/effect of agricultural interventions; (4) availability and accessibility of full-length peerreviewed articles; and, (5) study design considerations. As a result, a total of 630 publications are obtained using the advanced search in the title field. After checking for duplicates and irrelevant contents, 284 papers are saved to Mendeley. From these, 163 relevant papers are found after reading the title, abstract, and keywords. A full-length quick scan resulted in 75 potentially suitable papers. Finally, after reading throughout these papers, 25 studies are selected for the review. 2

Description of the studies included in the review
The types of development interventions included in this review are given in Table 1. Geographically, we note a high concentration of studies in few regions of the country, i.e., Southern Nations, Nationalities and Peoples (SNNP), Amhara, Oromia, and Tigray. No studies are found for four regions (Afar, Benishangul Gumuz, Gambela, and Harari), Dire Dawa Administrative Council, and Addis Ababa City Administration. The main units of analysis in the reviewed studies are households (20%), children 6-24 months old (20%), children under five years of age (16%), and mother/ caregiver and children (16%).

Crop production 6
Livestock production and management (exclusive) b 6 Livestock (general) c 5 Productive safety net program (PSNP) 3 Nutrition education and complementary feeding 4 Agricultural water use 3 Agricultural production (general) d 4 a The numbers do not add up to 25 because some studies fell into more than one intervention category.
b These refer to interventions that focused entirely on livestock production and management. c These are studies that have many components but treated livestock production as one aspect.
Concerning study design, 15 out of the 25 studies were based on cross-sectional survey design; seven studies used quasi-experimental design; and, the rest employed longitudinal design. All the studies used questionnaires to gather data; 12 of them applied measurements (to obtain anthropometric data). Other methods, such as focused group discussions (FGDs), key-informant interviews (KIIs), and observations were rarely used. Data analysis was predominantly carried out through regression techniques (60%) and mean comparisons (20%). Other analytical procedures include propensity score matching (PSM) combined with regression (8%), PSM only (8%), and qualitative analysis (4%). We identified a total of 21 outcome/impact indicators, which we grouped into six broader categories following . 3

The effect of crop production on dietary diversity, consumption practices, and child anthropometry
The list of studies in the crop production category is presented in Table 2. We find that half of the studies measured dietary diversity in children in Amhara (Beyero et al., 2015; and SNNP (Dangura & Gebremedhin, 2017) regions. Overall, all the studies document a significant positive association between ownership of home/backyard garden and children's dietary diversity. In particular, possessing home gardens doubled the likelihood of children having minimum dietary diversity (Beyero et al., 2015). Dangura and Gebremedhin (2017) found that the dietary diversity score of such children is 0.32 higher than their counterparts in households without home gardens. Nonetheless, the relationship between home gardening and minimum meal frequency is found to be insignificant (Beyero et al., 2015).
Three out of the six studies assessed changes in consumption practices. The assessment of the orange-fleshed sweet potato (OFSP) intervention in the SNNP region revealed that although 63% of the sampled households grew potato, only 7% of them produced OFSP. Moreover, it was found that only 1% of the 63% of mothers consumed OFSP . Another study by Negash and Swinnen (2013) showed that castor bean contract farming in the SNNP region increased food consumption and reduced the length of food gap months.
Concerning child anthropometry,  showed that adoption of improved maize varieties (IMV) positively and significantly affected children's height-for-age Z-score (HAZ) and weight-for-age Z-score (WAZ) but not weight-for-height Z-score (WHZ) in four regional states of the country (i.e., Oromia, Amhara, SNNP, and Tigray). The empirical findings show that allocating 0.25 hectare of land for IMV cultivation resulted in a HAZ and WAZ increase of 0.257 and 0.176 standard deviations, respectively.  Source: Based on the author's elaboration of reviewed papers. Table 3 presents the six studies in this sub-category. It should be noted here that 50% of the studies (Geday et al., 2016; were based on quasi-experimental designs of impact evaluation, with "intervention" and "control" groups specifically identified. Moreover, these studies employed the propensity score matching (PSM) technique in their analysis.

Livestock production and management (a) Exclusive interventions in livestock production and management
Four out of the six studies evaluated the impact on anthropometric indices. Starting with stunting (HAZ), we find that all the studies provide evidence of a positive and significant reduction in stunting levels. For instance,  showed that owning a single cow has the potential to increase HAZ (between 0.25 and 0.47 standard deviations) and reduce stunting levels (between 6% and 13%). Positive associations between livestock ownership and HAZ were also documented by Headey et al. (2017) and Headey and Hirvonen (2016).  also found a significant mean difference between non-participants and participants of the formal milk market. Concerning wasting (WHZ), three out of the four studies measured it but found mixed results. Whereas  and  found no significant impact, Headey et al. (2017) documented a positive association between livestock ownership and WHZ.  also found improvements in underweight (weight-for-age Z-score-WAZ) for children in market participant households, although the estimated mean difference between participants and non-participants is significant only with Kernel matching (and not with Nearest Neighbour matching).
Although the above studies provide empirical evidence that livestock ownership has the potential to improve nutrition, such a generalization is not always warranted. For instance, exposure to livestock excreta was shown to negatively affect HAZ and WHZ in Ethiopia (Headey et al. 2017). Likewise, keeping poultry in the same house with people is found to be negatively associated with HAZ (Headey and Hirvonen, 2016).
Two of the six studies analyzed the causal effect on household income.  found that adopters of dairy development interventions (i.e., cross-bred dairy cows and improved forage) earned a significantly higher income than non-adopters. The milk value-chain participation study by Geday et al. (2016) also found a positive and significant gain in household income, although the Kernel matching result was not significant, casting doubts on the robustness of the impact estimates.
A positive and significant effect was also found for dietary diversity in three out of six studies. A unique feature of the study by  is that it provides, in addition to the household dietary diversity, an analysis result of intra-household dietary diversity. In addition to dietary diversity, Geday et al. (2016) also showed that per-capita calorie intake was higher for participants. Nevertheless, similar to our earlier reservation, the average treatment effect (ATT) should also be treated with caution for the same reason. Some of the studies investigated the impact on production, consumption, and marketing of animal source foods (ASFs). The evaluation by  showed that milk production was higher for market participant households. Concerning consumption, however, there are mixed findings. Whereas  found that cow ownership increases the probability and frequency of milk consumption, Geday et al. (2016) showed that participants of the milk value chain consumed less milk. On the contrary,  documented no significant difference between milk market participants and non-participants. There are also mixed findings on the consumption of other ASFs. Despite the findings that egg consumption by children increased, no evidence was found for increased consumption of meat and other dairy products due to poultry production (Headey and Hirvonen 2016).  demonstrated that cow ownership significantly increases dairy product consumption. In relation to marketing,  found that milk market participants sold higher quantities of milk, which may be the reason why such households had a lower level of milk consumption.
Only one of the six studies assessed the effect of livestock production on child illness and maternal/child cleanliness (Headey et al., 2017). The results indicate a strongly negative association between the presence of animal excreta around the homestead and maternal/child cleanliness. However, the presence of animal faeces or ownership of livestock was found to be not associated with illness (e.g., diarrhoea, fever, and cough/cold).

(b) Other studies with a livestock component
The list of studies that have a livestock component is depicted in Table 4. The major focus of these studies is dietary diversity. It has to be noted, however, that none of these studies are strictly impact evaluations. In general, all the studies found a positive association between livestock ownership and dietary diversity. Specifically, it is noteworthy that although a huge rural-urban gap in children's dietary diversity was revealed, the higher probability of owning livestock in rural areas contributed to narrowing down this difference . Concerning diet quantity, Goshu et al. (2013) found no effect on daily calorie availability.

The Productive Safety Net Program (PSNP) and its impact on food security, asset holding, and child nutrition
Three studies investigated the impact of Ethiopia's Productive Safety Net Program (PSNP) on food and nutrition security (Table 5). In a relatively large study based on longitudinal data obtained from four regions of the country (i.e., Oromia, Amhara, SNNP, and Tigray),  estimated impact of the PSNP, and that of other food security programs (OFSP) and household asset building programs (HABP) operating in the study regions. In doing so, the study not only accounted for some potential confounders but also quantified the combined impact of the PSNP and these programs. A key finding of the study is that participation in the PSNP improved food security and raised livestock holdings. The combined impact of the PSNP and OFSP/HABP is shown to be greater than the impact of each intervention separately. Table 4. Characteristics of other studies in livestock production and management

Study
Study unit Sample size Goshu et al. (2013) Households 260 Hirvonen ( The findings of  in Tigray region revealed that children in PSNP member households have WHZ that were higher than those of children in non-member households. In another PSNP evaluation, Baye et al. (2014) documented that the dietary diversity of households receiving food was significantly lower than that of households receiving cash. Households receiving cash had significantly higher levels of consumption of oils and fats, and somewhat higher consumption of legumes and dairy products. Although statistically insignificant, the average HAZ and the prevalence of stunting were also lower in households receiving cash.

Nutrition education/complementary feeding interventions
The first three studies in Table 6 assessed the mediating role of nutrition education interventions in promoting pulse-based complementary feeding practices in the SNNP region.
One of the major findings of Yetnayet et al. (2017) relates to knowledge, attitude, and practice (KAP) of food-based approaches to nutrient deficiency. In this regard, the study found significant changes in the KAP scores of the participants due to exposure to the intervention. Concerning consumption, the study showed substantial improvements in the proportion and consumption frequency of pulse-based diets in the intervention group. Similar changes in knowledge and practice of complementary feeding were documented by . However, unlike Yetnayet et al. (2017), they did not measure changes in attitude. Analogous to the findings in Yetnayet et al. (2017), there is clear evidence on the enhanced consumption of legumes as a result of the intervention. An added dimension in this study, unlike Yetnayet et al. (2017), is the analysis pertaining to anthropometric indices, although no significant changes in height and weight of children were detected.
The findings reported in  are also the same as those found in Yetnayet et al. (2017) for KAP scores and consumption practices. They are also related to  in terms of knowledge and practice scores. However, contrary to ,  found a positive and significant effect of the intervention on WAZ and WHZ. However, the effect of the intervention on HAZ was found to be negative. The study by  investigated the role of caregiver's nutrition knowledge on dietary diversity of children in north-west Ethiopia and found a positive and considerable effect. However, the study also illustrated that the impact of nutrition knowledge decreases as one moves far away from the Source: Based on the author's elaboration of reviewed papers.

Agricultural water use (irrigation) interventions
There are three studies on agricultural water use (Table 7). Surprisingly, the observational study in Central and Hararghe highlands found that irrigation water use adversely affects the household's daily calorie availability, diet diversity, and food security status (Goshu et al., 2013). In a stark contrast, however,  showed a positive effect of small-scale irrigation on food security in Amhara region. This study also found higher crop diversity and better nutritional status in villages where irrigation is practiced. Moreover, crop loss was minimal and drought-tolerance capacity was better in these villages.
The study by Hagos et al. (2017) also provides evidence for a positive effect of irrigation on the depth and severity of poverty. The study found a statistically significant difference between spate irrigation users and non-users in terms of absolute poverty, poverty gap, and severity of absolute poverty. The comparison between traditional and modern spate irrigation schemes indicated that the use of modern spate irrigation resulted in a significant reduction of household poverty levels.

Effect of general agriculture interventions on dietary diversity, food security, and nutritional status
The studies investigating the effect of food production and access to farm resources (i.e., general agriculture) are given in Table 8. We note that all the studies are not strictly impacted evaluations. Two of the studies found that dietary diversity is enhanced through improved access to farm resources (i.e., fertilizer in Goshu et al., 2013) that could result in production diversity. Hirvonen and Hoddinott (2017) showed that a one food group increase in household's production diversity leads to a 0.49 (Poisson regression) to 0.62 (OLS regression) food group improvement in child's dietary diversity score.
The two studies that assessed food security found that access to farmland  and the use of fertilizer (Goshu et al., 2013) had a positive effect on food security status. In addition, Goshu et al. (2013) found that fertilizer use enhanced daily calorie availability and intake. However, the studies investigating nutritional status reported conflicting results. Whereas  documented that access to farmland plays a significant role in household nutritional status,  found higher odds of stunting among children whose parents use their own production as a source of food.

Impact pathways from agriculture to improved nutrition
In the 25 studies reviewed, we find varying levels of evidence for 6 impact pathways. No evidence is found for agricultural interventions affecting nutrition through women's enhanced control over resources. Overall, the most dominant mechanism linking agriculture to improved nutrition outcomes is through the production of crops to increase household consumption. This is followed by the income effect of agriculture. We find a very meagre support for the role of agriculture in enhancing non-food expenditures, and the nutrition and health of the household (Figure 1).
In Table 9, we present the proportion of the key pathways by intervention category. Although some pathways are common across the interventions, there are also a few of them uniquely important in only some intervention categories (see, for instance, "women's time and caring practices" in the PSNP and nutrition education interventions). In general, in six out of seven categories, we find more than three mechanisms from agriculture to nutrition. In what follows, we present discussions by intervention category.

Impact pathways in crop production
Overall, in all the studies in crop production, we find overwhelming evidence that crop production contributed to household's own consumption by serving as a source of food. For instance,  showed that improved maize cultivation raises household own-produced maize consumption that translates to increments in HAZ, WAZ, and WHZ. Likewise,  indicated that households with home gardens are twice as likely to provide diversified food to their children. Beyero et al. (2015) found that most farmers possessing home gardens used the produce for home consumption. Likewise, Dangura and Gebremedhin (2017) documented that many children in household's producing fruits and vegetables consumed vitamin A-rich foods. Finally, Negash and Swinnen (2013) illustrated that the production of cash crop (castor bean) helped farmers store their food crops to be consumed during the lean season.
Regarding the income effect of crop production, we find evidence only in one study (Negash and Swinnen 2013). In that study, it was shown that households earn income by selling castor beans. Such contract farming schemes also safeguarded farmers against low food crop prices during harvest because the farmers sell their castor beans to generate income, but store their food crops to be consumed or sold later. Hence, the households involved in the contract farming did not have to pay higher prices for food during the lean season. The OFSP study by  suggests that women's own health and nutrition could be improved through enhanced production and consumption of OFSP.

Impact pathways in livestock production and management
As can be seen in Table 9, the most prominent mechanisms in this category are through increased consumption and income. Concerning the consumption of animal source foods (ASFs),  found that cow ownership increases the likelihood and frequency of consuming milk, cheese, and yoghurt. Similarly, Headey and Hirvonen (2016) showed that poultry ownership increases the probability of egg consumption, but not the consumption of other ASFs. Contrary to  regarding milk consumption, Geday et al. (2016) found a negative relationship between production and household consumption due to the fact that most of the milk produced was sold.  found no significant difference in terms of milk consumption for the same reason. Similar to Headey and Hirvonen (2016), however,  also found no significant difference in terms of consumption of other ASFs. Suggestive evidence about enhanced consumption is also found in  and Headey et al. (2017).
Concerning the income effect of livestock interventions, we find evidence in four out of six studies. For instance,  found that adoption of cross-bred dairy cows and improved forages resulted in increased income;  showed that milk market participants have higher income that helped them boost their dietary diversity; and, Geday et al. (2016) indicated that higher incomes resulting from dairy value-chain participation may lead to consumption of better or diversified foods.
In relation to food prices/market, we find evidence only in two studies. Whereas Geday et al. (2016) indicated that higher income from milk value-chain participation can improve the ability of households to buy nutritious foods from the market (i.e., food expenditure),  argued that cow ownership is less relevant in situations where households have better access to markets. The study by Geday et al. (2016) also provided indicative evidence that higher income due to milk market participation can improve non-food expenditures.
Two studies indicated that exposure to livestock excreta can have adverse effects on nutrition and health. In particular, Headey and Hirvonen (2016) showed that indoor keeping of poultry is negatively associated with children's HAZ. Likewise, Headey et al. (2017) found that livestock ownership and poor personal cleanliness are strongly associated with the presence of animal excreta. Finally, among the five general livestock-related studies, we find evidence for consumption in three studies (Goshu et al., 2013;. Goshu et al. (2013) and  also contain some proof for income effect.  and  provide support for food price/market effect. Lastly,  showed that nutrition knowledge of caregivers positively affects children's dietary diversity through enhanced caring practices.

Impact pathways in nutrition education and complementary feeding interventions
As can be seen in Table 9, the two dominant pathways in this category are improved consumption and caring practices. Three out of four interventions provide evidence that nutrition education sessions enhanced consumption of locally available pulse-based diets by women and children Yetnayet et al., 2017). Additionally, we find evidence on enhanced caregivers' knowledge and practices regarding complementary feeding . Nevertheless, as  caution, better nutrition knowledge translates to substantial improvements in children's dietary diversity only in areas with relatively good market access, or better own production of nutritious foods. The market-related result complements the findings of  regarding the importance of integrating smallholders to well-functioning markets. Finally, it can be deduced from Yetnayet et al. (2017) that increased consumption of pulse-based foods can stimulate women's own nutrition and health.

Impact pathways in the Productive Safety Net Program (PSNP)
In all the PSNP interventions, we find strong evidence for enhanced consumption, income, and food price/market effects. The food-for-work component of the PSNP is one pathway to enhance household consumption. The cash transfer component of the PSNP works to improve household income that translates to, for instance, higher consumption of ASFs and vitamin A-rich plant foods (Baye et al., 2014). The PSNP also generally improved the household's purchasing power despite the rising food prices  that adversely affect beneficiaries of the cash transfer component of the PSNP (Baye et al., 2014). Finally, since most Public Works activities of the PSNP occur in the months between January and June , extra female labour can be mobilized for such activities  without affecting their time for farming and caring practices.

Impact pathways in agricultural water use (irrigation) interventions
In all the studies in agricultural water use, we find evidence for consumption and income effects. Generally, irrigation helped households produce diversified crops. However, Goshu et al. (2013) argue that the focus of crop production in irrigated agriculture may shift from staple food crop to cash crop production. This, in turn, may result in increased household income Hagos et al., 2017). Hagos et al. (2017) also showed that food and non-food expenditures were significantly higher for irrigation users, although the items purchased did not appear to have contributed to improved nutrition.

Impact pathways in general agriculture
All the studies in this category provide evidence for the consumption effect. For instance, access to farmland enhances crop diversity , and increased crop diversity leads to improved children's dietary diversity (Hirvonen and Hoddinott, 2017). Contrary to these, own production as a source of food is found to be significantly associated with higher odds of stunting probably due to low productivity and poor complementary feeding practices . Regarding income effect, Goshu et al. (2013) showed that an increase in crop income reduced total income inequality. Finally, Hirvonen and Hoddinott (2017) illustrated that market access plays a greater role than production diversity in children's dietary diversity.

Conclusions and policy implications
In the present study, we investigate the impact of agricultural interventions and discuss the pathways linking agriculture to improved food and nutrition outcomes. Overall, we find positive effects on knowledge, attitude, and practice (KAP) and economic indicators (income, poverty and asset accumulation). We also find an overwhelming positive effect on food consumption, dietary diversity, and other food security indicators. However, we also note mixed effectiveness pertaining to anthropometric indices. 4 In addition, the most important pathways linking agriculture to improved nutrition center on its role as source of food, source of income, and food prices/market. Consequently, the following are lessons and recommendations for practitioners, planners and implementers of nutrition-sensitive interventions, and policy makers.
To start with, sustained community-based nutrition education interventions in diverse agroecological settings are important to accelerate the adoption of improved technologies and best practices, and promote household's consumption of nutrient-dense foods. Such efforts should be complemented by strategies to improve access to farm inputs, physical and financial resources, technical information, and input/output markets. Social protection schemes, such as the PSNP, should be integrated with other rural development interventions for maximum effectiveness and sustainability of impact.
The design, implementation, and evaluation of such interventions should take into account several issues. First, increased coordination among multisectoral actors (especially agricultural education, research and extension; health and nutrition) is required to build capacities at various levels. Second, the interventions should be designed and implemented in such a way that they contribute to improvements in nutrition and health outcomes of individuals, households and communities. In this regard, there should be a concerted effort to improve methodological approaches: include diverse agricultural, health and nutrition indicators; collect pre-intervention/baseline data; implement (quasi-) experimental designs in treatment allocation; gather time-series/panel data that can capture seasonality of production and consumption; and, employ rigorous impact evaluation techniques that account for unobserved heterogeneity, spillover effects, and sensitivity of parameter estimates to different specifications. Moreover, there should be a special attention on more qualitative studies and mixed research designs in order to capture evolutionary dynamics in food habits and consumption patterns and preferences. Finally, future research should also investigate, in addition to the abovementioned issues, the impact of policies and governance structures and research and development expenditures in relation to nutrition-sensitive agricultural interventions by employing meta-analysis techniques.