Participatory farm diversification and nutrition education increase dietary diversity in Western Kenya

Abstract Our study assessed the effectiveness of a community‐based participatory approach in increasing micronutrient adequacy of diets of women and young children through agricultural activities and nutrition education in Vihiga County, Western Kenya. Outcome indicators include the mean dietary diversity score (DDS), the percentage of women and children reaching minimum dietary diversity (MDD), and micronutrient adequacy (mean adequacy ratio). The project consisted of(a) a diagnostic survey covering agrobiodiversity and nutrition, (b) participatory development of activities to improve nutrition, (c) a baseline survey covering dietary intakes, (d) participatory implementation of the developed activities, and (e) an endline survey covering dietary intakes. The diagnostic survey was conducted in 10 sublocations of Vihiga County, which were pair‐matched and split into five intervention and five control sublocations. The intervention sublocations developed activities towards improving nutrition. Before implementation, a baseline survey collected the dietary intake data of 330 women–child pairs in the intervention and control sublocations. To support the activities, communities received agriculture and nutrition training. After 1 year of implementation, an endline survey collected dietary intake data from 444 women–child pairs in the intervention and control sublocations. Impact was assessed using the difference‐in‐difference technique. Highly significant positive impacts on children's mean DDS (treatment effect = 0.7, p < 0.001) and on the share of children reaching MDD (treatment effect = 0.2, p < 0.001) were shown. Higher dietary diversity can be explained by the development of subsistence and income‐generating pathways and increased nutrition knowledge. Participatory farm diversification and nutrition education were shown to significantly increase dietary diversity of young children in Western Kenya.


| INTRODUCTION
Malnutrition is a global challenge with huge social and economic costs, and constitutes the largest risk factor for the global burden of disease (IFPRI, 2016). The number of chronically undernourished people in the world reached 815 million in 2016; 155 million children were stunted, whereas almost 52 million children were wasted (FAO et al., 2017).
Multiple forms of malnutrition coexist, with countries experiencing simultaneously high rates of child undernutrition, anaemia amongst women, and adult obesity.
Participatory approaches are now widely accepted in development practice, also aimed at improving nutrition outcomes. Most of these studies focus on community participation during the implementation phase of an intervention. Faber, Witten, and Drimie (2011), Kang, Kim, Sinamo, and Christian (2016),  Kang et al. (2016) demonstrated that a positive deviance/hearth approach (Bullen, 2011; community-based rehabilitation and behaviour change intervention for families with underweight preschool children) combined with existing government nutrition programmes, could effectively improve child growth in rural Ethiopia. Studies using community participation at the intervention development level are, however, rare.
There is a research gap on the evidence of improved nutrition outcomes in interventions that apply a community participation approach throughout the project. This means that communities would decide for themselves about the intervention activities they will implement. Participation thus occurs not only at the intervention implementation level but is initiated at the intervention development (identification and planning) stage. Our study contributes to filling this research gap and adds knowledge on a model for community participation in the field of nutrition, both at intervention development and implementation level. Our research combined formal scientific methods, by conducting a quasi-experimental study and applying well-established indicators (e.g., dietary diversity score [DDS], mean adequacy ratio [MAR]), with participatory action (interventions were identified, planned, and implemented by the community). Furthermore, communities and local partners were involved in data collection and monitoring, and were invited to reflect on the study results. The combination of these approaches matches the definition of "participatory action research" (Baum, MacDougall, & Smith, 2006  In August and September 2015, 36 men and women per intervention sublocation (180 in total) were selected, with the help of community health volunteers (CHVs), to participate in a series of six workshops. One third of workshop participation was reserved for women with a young child, the second third for male farmers, and the last third for community members whose decision-making role can affect childcare and nutrition decisions (village elders, spiritual leaders, teachers, etc.). The CHVs were well known to selected community members, which served to ensure a level of trust from the participants.

| Study design and sample size
Minimum required sample sizes for the baseline and endline survey were calculated based on a formula proposed by Magnani (1999). baseline, and endline, the same number of HHs was selected per sublocation. In both baseline and endline survey, we randomly sampled two groups (intervention and control group) with 165 participating HHs in each. In addition, for the endline survey, we also sampled all women with a young child who could prove (through a membership FIGURE 1 Study tmeline list) after 1 year of implementation that they are project members.
These project members are referred to as the group of "direct beneficiaries." The vast majority of them had already participated in the community workshops prior to the implementation of activities, and some of them joined although the implementation was already ongoing. The community members of the same sublocations in the intervention group who did not join the workshops and implementation of activities are referred to as the group of "indirect beneficiaries." We therefore had a total sample size of 498 HHs (165 indirect beneficiaries, 165 control group members, and 168 direct beneficiaries) at endline.
The baseline survey was carried out in November 2015 (plenty season), before community activities started. Comprehensive lists of HHs with a young child aged 12-23 months were generated within the five intervention and the five control sublocations with help of CHVs, village elders, chiefs, and assistant chiefs. A total of 330 HHs (33 HHs per sublocation; 165 HHs in intervention and 165 HHs in control sublocations) were randomly selected using the RAND MS excel function. Dietary intake data were collected for a woman of reproductive age (15-49 years) and young child (12-23 months) in each of the selected HHs. In most cases, the woman was the biological mother of the child. We chose these target groups because of the critical consequences of poor nutrition during pregnancy and in the first 2 years of life can have on health and development throughout the course of life (Black et al., 2013).
One year later, in November 2016 (plenty season), we conducted an endline survey in the same 10 sublocations. As some of the children had grown out of the defined age bracket, the lists of HHs were updated, so that different HHs were interviewed at endline. We

FIGURE 2
Flow diagram for the baseline (2015) and endline survey (2016) therefore randomly sampled 330 women-child pairs from the lists of HHs with a 6-23 month-old child (indirect beneficiaries 165; control 165). Additionally, we interviewed all women with a young child in the group of direct beneficiaries (168). As in the baseline survey, dietary intake data were collected. For data analysis, the groups of direct and indirect beneficiaries were combined in the intervention group.
Having the group of direct and indirect beneficiaries was supposed to enable the assessment of potential spill-over effects from project participants (direct beneficiaries) to other community members (indirect beneficiaries). Figure 2 shows a flow diagram for the baseline (2015) and endline survey (2016).

| The intervention
The workshops were designed to encourage and support communities in autonomously identifying and planning agricultural activities to improve nutrition, as well as raising awareness on nutrition, and to dis-

| Outcomes
The The mean adequacy ratio (MAR) was determined by the nutrient adequacy ratio (NAR) (WHO, 2002). The NAR was calculated for energy, macronutrients (carbohydrates, proteins, and fats), and 11 micronutrients (niacin, thiamin, riboflavin, vitamin B6, vitamin B12, vitamin C, vitamin A, folate, iron, zinc, and calcium) as the ratio of the subject's nutrient intake to the estimated average requirements.
The estimated average requirement values were adjusted for the specific needs of pregnant women and children under 2 years. The lowest bioavailability for zinc (15%) and iron (5%) were used (WHO, FAO, 2006) as the recorded diets were predominantly plant-based (Gibson & Ferguson, 2008). The MAR was used to assess the overall micronutrient quality of the diet and was calculated as the sum of NAR for each of the micronutrients divided by the number of micronutrients considered (in this case, 11). NAR values were truncated at 100%.

| Data collection tools and procedures
The daily food intake of young children and their caregivers was assessed by the repeated, non-consecutive quantitative 24-hour food intake recall method. All foods and beverages consumed during the preceding 24 hours, including ingredients and cooking methods of mixed dishes, were listed. The amounts of all foods, beverages, and ingredients of mixed dishes prepared were estimated either in weight, HH units (volume determined by water content), or in monetary value.
The proportion of what was eaten was determined based on the volume eaten and the total volume of the prepared dish. This proportion was used to calculate the amount of ingredients consumed. For dishes consumed outside the home, standard recipes were prepared, and the amount of ingredients consumed by the subject was determined. To convert monetary values to respective weights, a market survey was conducted in each sublocation. A total of three weights of the edible portion were recorded for each food item from at least three vendors in the market and a mean obtained. Conversion factors from HH measures and monetary values to weight equivalents were then calculated (Gibson & Ferguson, 2008 (Sehmi, 1993), and the West African food composition table (Stadlmayr et al., 2012). Nutrient composition of raw ingredients was corrected for loss of nutrients during cooking using the United States Department of Agriculture

| Data management and statistical analysis
Food intake data from the two 24-hour recalls were entered and processed in Lucille analysis software. Usual food group and nutrient intake distributions were generated using the multiple source method (EFCOVAL, 2010;Haubrock et al., 2011). This method allows elimination of intrapersonal variation of the intake of the nutrient/food group.
Statistical analysis was done using IBM SPSS Statistics Base, version 22. Means were compared using t tests and one-way analysis of variance, whereas proportions were compared using the Pearson chisquare. We used a 95% confidence level in all cases. The project's causal impact, thus the effect of the community-based participatory approach on dietary diversity was assessed using the difference-indifference (DID) technique inside a mixed effect multiple linear regression. Treatment and time were included in the covariates and sublocation was considered as a random factor. DID is used to estimate the effect of a specific intervention by comparing the changes in outcomes over time between an intervention and a control group (WHO, 2011). This applies to the sampling of a changing population at different points in time (Wooldridge, 2010), as was the case for the present study. Effect size refers to mean differences-indifferences that have not been standardised.

| RESULTS
Characteristics of the women-child pairs are presented in Table 1 Table 4).

| DISCUSSION
Our impact analysis showed a significant, positive impact of the intervention on the children's mean DDS and on the share of children reaching MDD. No significant impact was found on the women's    Table 2 shows that child DDS increased by about one food group (from 3.6 to 4.5 food groups). We consider the addition of one food group out of seven food groups in total as a meaningful improvement for the children's diets. The share of children consuming dairy products and flesh foods increased by 52% (from 61.1% to 92.6%) and 76% (21.0% to 37.0%), respectively, which we also consider a substantial increase. Furthermore, the share of children reaching MDD increased by 74% (from 50.9% to 88.7%), which we consider a noteworthy increase in the percentage of children who consume a diverse diet at population level. The inclusion of variables related to the women's and children's diet (mother/caregiver's age, child's age, mother/caregiver's educational status, and mother/caregiver's pregnancy status) did not lead to significant changes in the results of the intervention group and direct beneficiaries. Regarding child DDS and child MDD in the indirect beneficiaries, the p values increased (from p = 0.003 to p = 0.813 and from p = 0.003 to p = 0.864, respectively).

Improved nutrition outcomes have also been shown in other
projects (see below), even though not necessarily participatory, that applied nutrition education and/or home-based food production approaches. There is growing interest in the potential of home-based food production to address micronutrient undernutrition in developing countries (e.g., Keatinge et al., 2012;Olney, Pedehombga, Ruel, & Dillon, 2015;Weinberger, 2013). Home gardens can be a useful food-based strategy to promote more balanced diets amongst poor rural HHs that have access to a small plot of land and are willing to engage in gardening (Schreinemachers, Patalagsaand, & Uddin, 2016). Studies conducted in Bangladesh, Cambodia, Nepal, and the Philippines found that families who participated in homestead food production activities benefited from increased production and consumption of vegetables, fruits, and poultry products (Helen Keller International, 2010).
A similar study from Nepal (Osei et al., 2016)   combination of them: increased accessibility to green leafy vegetables and legumes due to increased home-based production of these crops, increased accessibility to other foods due to an increase in income (by selling kitchen garden crops and/or chicken and eggs), and increased nutrition knowledge gained through the nutrition education (that focused on the importance of a diverse diet). Increased dietary diversity in the indirect beneficiaries can be related to dissemination regarding nutritional and agricultural knowledge. This again may have led to increased income and diversified agricultural production (as in the direct beneficiaries). It could also be explained by the nutrition education that reached almost half of the indirect beneficiaries.
It might be easier for mothers to adapt their young children's diet, rather than adapting their own or the HH adults' diet, as adult dietary habits are already more firmly developed. We therefore assume that the HH adults need longer than 1 year to significantly improve their The participatory nature, which implied participation throughout the project, is the strength of this study. The level of community participation in this project, measured by the model of Kc et al. (2011) that defines different degrees of community participation assessed in terms of ownership and sustainability, can be set at 5 on a scale from 1 to 6. This is the second highest degree of community-participation, defined as "co-learning: local people and outsiders share their knowledge to create new understanding and work together to form action plans with outsider facilitation." To reach Level 6, local people would have needed to carry out their agenda in the absence of outside initiators and facilitation, which only partially occurred.
We assume that the participatory nature of the project significantly contributed to the improvements in outcome indicators. We observed that trust and group dynamics were built amongst the direct beneficiaries during the 2-month workshop period and that trust was built between them and the researchers. It took a few workshops before the direct beneficiaries understood that they were the main actors of the workshops and for them to gain ownership over their own activities, because from previous projects, led by other organisations, they were used to following rather than developing and This study is a good example of a research for development project that implements a participatory approach with an initially smaller number of beneficiaries, but that reaches a much higher number due to dissemination and government uptake. This is one of the additional benefits of a participatory approach, apart from improving outcome indicators. However, we still need to assess why this approach was highly accepted and easily adopted by the community, calling upon social sciences studies to evaluate the role of group and gender dynamics, amongst others. As do several other studies (Newig & Fritsch, 2009;von Korf, Daniell, Moellenkamp, Bots, & Bijlsma, 2012;Munang & Nkem, 2009), we support the premise that community participation leads to more sustained and better decision uptake. As interventions are still ongoing (under a different donor since 2017 and with a focus on guiding the communities into farmer resource centres), we have been able to verify that the kitchen-gardening and poultry-raising activities are still very much on-going even though related trainings have stopped. Another strength of our study includes the possibility of assessing spill-over effects into non-participating communities.

| CONCLUSIONS
A community-based participatory approach was applied to one group in five different communities. Nutrition outcomes of this approach were then assessed for the five communities (not only for the groups), clearly showing that participatory farm diversification and nutrition education significantly increased dietary diversity of young children in Western Kenya. However, there was no significant impact on women's nutrition outcomes. This project contributed to filling the research gap on the evidence of improved nutrition outcomes through an approach that applies community participation throughout. The positive results can inspire future nutrition-related projects to expand their participatory component in a way that allows understanding of the communities' context, allows the communities to gain ownership from the outset, and decide autonomously how to create changes in their environment.
It would be important to assess long-term benefits that a participatory approach might provide, such as group cohesion, close partnerships between the beneficiaries and local agriculture and health (extension) workers, policy makers and NGOs; women's empowerment and spillover effects. For scaling-out activities (within and outside Kenya), it will be crucial to know more about the determinants for community participation and high acceptance. We will analyse this in a separate paper.

ACKNOWLEDGMENTS
We thank the CGIAR Research Programmes HumidTropics for financing the field work and Agriculture for Nutrition and Health for the cofinancing. We also thank GIZ and Biovision for supporting research staff positions. Special thanks go to the community health and agricultural extension workers and nutritionists of Vihiga County as well as to the NGO WeRATE and the enumerators. We also thank Olga Spellman (Bioversity International) for language editing.

CONFLICTS OF INTEREST
The authors declare that they have no conflict of interest.

CONTRIBUTIONS
CT, JB, FOO conceptualised the scope and framework of the study.
FOO, JB and CL conducted the statistical analysis for this article. All authors contributed to writing.

RELEVANCE OF MANUSCRIPT
Participatory approaches are now widely accepted in development practice and also in the attempt to improve nutrition outcomes. There is, however, a research gap on the evidence of improved nutrition outcomes through an approach that implies community participation throughout the project. This means that communities decide themselves about the intervention they will implement. Participation thus happens not only at the intervention implementation level but already at the intervention development (identification and planning) level.
Our study contributes to fill this research gap and adds knowledge on how community participation in the field of nutrition can look like, both at intervention development and implementation level.