Can governments promote homestead gardening at scale? Evidence from Ethiopia

Low intake of fruits and vegetables is a major cause of micronutrient deficiencies in the developing world. Since the 1980s, various non-governmental organizations have promoted homestead gardening (HG) programs, first in Asia, but now increasingly in Africa. Longstanding concerns with HG programs are: (1) they lack scalability, particularly for governments; (2) they only work in areas with/without good access to markets; and (3) they are only suitable for more water-abundant ecologies. We assess these concerns by analyzing a large and novel survey on the adoption of a nationwide HG program implemented by the Ethiopian government. We find that better market access encourages HG adoption; so too does greater public promotion of HGs, but only in more water-abundant ecologies. This study focuses on the four most populous highland regions of Ethiopia; Amhara, Oromia, Southern Nations, Nationalities, and Peoples’ Region (SNNP) and Tigray. These regions are largely covered by mountains and elevated plateaus that together cover about two thirds of the country and host more than 85% of the total population (CSA, 2013). Agricultural production is largely rain-fed and dominated by cereals and pulses (Bachewe et al., 2017). This study is further constrained to areas within these regions in which the Productive Safety Net Program (PSNP) operates. With 8 million beneficiaries, PSNP is one of the largest safety net programs in Africa. Geographically, the program is targeted to chronically food insecure districts (woredas). Most PSNP beneficiary households receive cash or food (mostly in the form of cereals) payments for undertaking public works while a small proportion of households with limited labour capacity receive unconditional payments. Food security in these localities has been improving but remains high. The data underlying this paper show that the prevalence of stunting among children 6–23 months of age is high at 39%. Moreover, diets are extremely monotonous; the mean dietary diversity based on 24 h recall among children 6–23 months was 2.8 food groups (out of 7) and among mothers 2.2 food groups (out of 10). Less than 10% of the children (6–23 months) and mothers reported to have consumed Vitamin A rich fruits or vegetables in the past 24 h. Homestead gardening has been a part and parcel of tropical food production for millennia (Kumar and Nair, 2004). This is also true in Ethiopia (McCann, 1995), although most small-scale “backyard” production still focuses on staple crops (enset, maize and teff) or stimulants (coffee, khat) (Mellisse et al., 2017), rather than nutrient-rich FVs. The Ethiopian government therefore seeks to promote small-scale FV production as an important tool to increase availability of FVs at the household and community level. Both the National Nutrition Program and the Nutrition Sensitive Agriculture Strategy set explicit targets for HG adoption: 40% of rural households by 2020, and 25% of urban households by 2020 (GFDRE, 2016, MoANR, & MoLF, 2016). The government’s HG activities are implemented as part of a broader package of community-based nutrition-specific and nutrition-sensitive services, including BCC to promote age appropriate feeding practices, improved growth monitoring, treatment of severe acute malnutrition, disease prevention and management, social safety nets, and other agricultural interventions (FDRE, 2016, GFDRE, 2016). Many of the nutrition-specific activities are implemented by health extension workers (HEWs), who are also tasked with promoting household adoption of HGs. Community volunteers, known as the Health Development Army, assist HEWs in implementing these services, and agricultural extension workers (AEWs) are further tasked with providing technical support to households that wish to adopt. So far, HG activities have focused on promotion and technical support. The provision of inputs such as seeds and seedlings for HGs by the HEWs or AEWs has been rare. Moreover, the production of poultry or other small ruminants is not yet widely promoted either. This study uses three types of data: (1) secondary household survey data collected by the authors in March and August in 2017 in PSNP districts in Amhara, Oromia, SNNP and Tigray; (2) community level data collected from HEWs, as well as from community food markets; and (3) Geographic Information Systems (GIS) data on agro-climatic factors. The original purpose of these surveys was to obtain information for an evaluation of nutrition sensitive components of the PSNP. To this end, a stratified sample was drawn from areas in which the PSNP operates in the four highland regions. Given the focus of the original evaluation on outcomes related to child nutrition, the sample was restricted to poor households with a child less than 24 months of age in March 2017. Supplemental File S1 describes the sampling strategy in more detail. While the geographic and demographic restriction is potentially a limitation on external validity, the survey has several useful characteristics. First, this survey is longitudinal and designed to capture seasonal differences. The first survey round was administered 3–4 months after the main harvest season in March 2017 (N = 2635), in what is typically a dry season. A second follow-up survey of the same households was conducted approximately six months later in August 2017 (N = 2569) during the long rainy season (2.5% of first round participants were not available in the August 2017 round). Second, the sample is large and geographically extensive (see Fig. 1), covering 264 enumeration areas (EA) from 264 sub-districts (kebele) and 88 districts (woreda). Moreover, although all 264 communities are poor, they vary substantially in terms of agro-climatic conditions, population density and access to markets. Lastly, the survey collected data on a wide range of household demographic and socioeconomic characteristics, including asset ownership and maternal nutrition knowledge (with the primary respondent being the mother of the young child). Locations of the 264 communities in the household survey and their overlap with average rainfall over 1997–2015. The field team also interviewed 249 HEWs working in the same localities in the August 2017 round, including questions on HGs. In addition, the nearest food markets of the communities (N = 264) were visited in both survey rounds. The market survey instrument recorded information on market characteristics (e.g. size of the market, infrastructure) and prices of 71 commonly consumed food items. For each sampled household and food market, we obtained latitude and longitude coordinates using Global Positioning System (GPS) devices. We used these data to compute the geodetic distance between the household and the nearest food market. We also linked the household level GIS data to daily weather data from National Aeronautics and Space Administration (NASA) and used these to compute the mean number of rainy days in the locality over 1997–2015. The location of the 264 communities across the four highland regions and their overlap with different rainfall brackets is depicted in Fig. 1. We used these different data sources to construct an array of indicators of HG adoption and its determinants (details for all variables used in the analysis are provided in Supplemental File S2). HGs were defined as a small area next to the house in which the household cultivates FVs. In each survey round (March and August), households were asked whether they had cultivated a homestead garden in the past 12 months. Based on the responses to this question, we created a binary variable that obtained one if the household reported to have cultivated a HG in the past 12 months in either March or August. Respondents were also asked why they did not adopt a HG, with specific responses for “poor access to water”, “not enough available land”, “not enough time”, “not having seeds/inputs”, “not having the skills/knowledge”, along with a generic “other reasons” option. The survey also asked whether HEWs, volunteer health workers or AEWs promoted HGs. We used these responses to measure community level promotion intensity of HGs in each locality by taking the share of other households in the EA that reported receiving encouragement from extension workers (i.e. we excluded the household’s own response when calculating the EA level promotion rate). Two different measures of water access were used: 1) median reported time in the EA to fetch water (to go to the water source, get water, and come back) during dry season and 2) mean number of rainy days in the locality. Market access was measured using the quality of the nearest food market and household’s distance to this market. To measure market quality, principal components analysis was used to construct an index from seven different market characteristics: Road accessibility; Road quality; Accessibility by public transportation (buses); Access to electricity; Access to cell phone network; Number of food traders; and Market type (permanent or temporary market). Further analysis (see Supplemental File S3) shows that physical accessibility of the market, size of the market and the availability of electricity are major determinants of this market quality index. The final market quality index used in the analysis varies in value between 0 and 10 with higher values indicating better market quality. Household’s distance to the nearest food market is based on geodetic distance and constructed using GPS coordinates of the household and the market. Maternal nutrition knowledge was measured using mothers’ responses to 12 questions regarding infant and young child feeding (IYCF) practices (see Supplemental File S4). The final maternal nutrition knowledge score used in the analysis varies in value between 0 and 10 with higher values indicating better knowledge. All statistical analyses were conducted using Stata, version 15.0 (StataCorp, Texas). The analysis proceeds in stages. We first use descriptive statistics to compare the characteristics of adopting and non-adopting households, along with two-sample t-tests of significant group differences. We then provide a regional bivariate analysis of households’ homestead garden adoption and access to water and food markets. Multivariate probit methods were then used to test whether the observed associations between HG adoption and various household characteristics remained significant after controlling for observable confounding factors. We also ran a sensitivity analysis in which communities were split between water-abundant and water-scarce areas, as determined by mean rainfall levels over 1997–2015. In all regression analyses, the standard errors were clustered at the EA level.