Intervention (Hidden)
SETTING AND CONTEXT
To assess the effect of video-mediated agricultural extension we compare how farmers respond to the same information regarding improved technologies and practices when this information is disseminated through the standard extension approach or when the approach is supported by video mediation.
As with any real-life policy, the video-mediated intervention was not limited to simply producing and screening videos for farmers. Rather, the intervention comprised three interlinked components: bringing farmers together in small development groups, conducting videos screening with development groups that were facilitated by extension agents, and verifying the uptake of the practices in the field.
Organizing farmers in small groups: Most farmers in Ethiopia are members of a development group, which is a semi-formal administrative structure within each kebele comprised of 25-30 farm households and designed as a grassroots forum for discussion of local development issues. The kebele is the primary level at which both administrative activities and agricultural extension is organized in Ethiopia. DAs assigned to a given kebele have access to these development groups, and thus use them as forums to introduce and discuss new agricultural technologies and practices. These technologies and practices are typically based on recommendations from MoA and the regional bureaus of agriculture.
Video-mediated discussions with extension agents: The cornerstone of Digital Green’s intervention is a video-mediated approach. Digital Green works with partners—woreda extension staff such as subject matter specialists and local NGOs where appropriate—to produce short videos featuring local farmers speaking in local languages about the subject technologies and practices. These videos were screened using USB-charged PICO projectors by local DAs assigned to the kebele. Videos were screened with development groups (or several development groups if necessary) in a manner designed to facilitate effective learning and discussion. Specifically, DAs would screen the videos several times during the meeting, and would pause the videos at certain intervals to entertain questions or provide additional details. DAs would augment their facilitation with input from model farmers belonging to the development group(s) present at the screening. These screening sessions would be conducted several times during the season in a manner that synchronized the video content with the crop calendar.
Adoption monitoring, and verification: The Digital Green approach is designed so that DAs and Digital Green staff can follow-up with farmers on their adoption of technologies and practices presented in the videos, either by querying farmers directly or verifying adoption visually. This follow-up feeds into Digital Green’s connect online-connect offline (COCO) system for project monitoring, evaluation, and learning.
EXPERIMENTAL DESIGN
This study uses a three-arm stratified cluster randomized controlled trial implemented in the four main regions in Ethiopia during the 2017 meher (rainy) season. Stratification was done at the level of the woreda. Clusters are defined at the kebele-level, which is the primary level at which agricultural extension is organized in Ethiopia. Within each woreda, kebeles were randomly allocated to one of three groups:
T0) A control group (denoted “Control”) in which the Government of Ethiopia’s conventional extension approach is targeted at the (typically male) household;
T1) A treatment group (denoted “Regular DG”) in which Digital Green’s standard video-mediated approach (described above) is targeted at the (typically male) household head; and
T2) A treatment group (denoted “DG + spouse”) in which Digital Green’s standard video-mediated approach is targeted at both the household head and his spouse.
In each group, the same suite of agricultural technologies and practices was promoted. In other words, Digital Green’s video-mediated approach did not affect the choice of technologies promoted through the Ethiopian extension system. The homogeneity of agricultural technologies promoted ensures that we are evaluating the medium used for promotion rather than the content that is being promoted. We focus on three main technologies (row planting, lower seeding rates, and application of fertilizers, specifically urea top dressing) promoted by the extension system for the three main cereals in Ethiopia (teff, wheat, and maize). By focusing on selected technologies and crops we ensure comparability of adoption rates and other outcomes within and across woredas.
This design allows us to estimate the impact of Digital Green’s video-mediated approach on several outcomes of interest including (i) awareness and understanding of specific agricultural technologies and practices; (ii) uptake of the technologies on one’s field in the ensuing agricultural season; and (iii) whether impact varies when extension is targeted to both male and female spouses in each participating household instead of the one, typically male, member per household.
SAMPLING STRATEGY
The study relied on Digital Green’s 2017 saturation plan in 68 woredas across the four main regions of Ethiopia—Amhara, Oromia, SNNP, and Tigray. The sample households were selected using a four-stage sampling process.
1. Defining the study population. In the first stage, we purposefully selected 30 woredas for the RCT based on three criteria: (i) woredas that were not saturated or fully covered by the Digital Green intervention prior to the 2017 meher season; (ii) woredas where Digital Green planned to expand to new kebeles; and (iii) woredas that would not be fully saturated during the 2017/18 expansion (to ensure the presence of within-woreda control kebeles). Woredas with less than nine potential expansion kebeles for the 2017/18 meher season were excluded from the study.
2. Stratification of the treatment at woreda level. In the second stage, we randomly selected kebeles from woredas with more than 15 eligible kebeles. Within each woreda, selected kebeles were randomly allocated to one of the three treatment arms such that each arm contained an equal number of kebeles.
3. Stratification of the sample by development group distance. Even though the kebele is the lowest administrative unit in Ethiopia, it typically comprises several development groups. Given the limited number of PICO projectors available for video screenings, it was usually infeasible for DAs to reach all development groups in a kebele with the video. Anecdotal evidence suggested that in such cases, DAs would likely focus their effort on the development groups close to the FTC. To assess the effectiveness of Digital Green’s approach on farmers, regardless of their location, we encouraged DAs to first focus their effort on 10 development groups—five of which were selected from the closest development groups (where distance to the FTC was less than the kebele median), and five from development groups located further away (where distance to the FTC was greater than the median).
4. Sampling farmers. For the last stage, we randomly selected seven households from each kebele: 2 from the closest targeted development group, 2 from the furthest development group, and 3 from the development group situated at the median distance from the FTC. Focusing on farmers within the ten development groups significantly increased the statistical power of the study. It ensured that a large share of the surveyed farmers targeted for treatment at the kebele-level, did in fact participate in it. As discussed later, the participation rate remains limited even under this sampling procedure.
Overall, the total sample comprises 30 woredas, 350 kebeles, and 2,422 farm households.
In Year 2 of the study (corresponding with the 2017 meher season), the intervention is repeated, and we revisit these households and DAs with the same survey instrument in early 2019. We also add an additional 900 households to the survey to capture spillover effects. These additional households reside in the same kebeles as the treatment households but are members of Development Groups that did not participate in the video-mediated extension approach.
Control group specificities. The selection of farmers to be surveyed followed the same procedure in both treatment and control kebeles. As shown below, this ensured comparability of farmers across groups. Our design sought not to affect, in any possible way, the way extension was carried out in kebeles in the control group. For this reason, we did not encourage DAs to focus their attention on ten development groups as we did in the treatment group kebeles. We discuss the implication of this difference below.
TIMELINE
We first conducted a baseline survey of all DAs working in our study kebeles during April and May 2017, which was right before the 2017/18 meher production season. The DA baseline survey collected information on their profiles, motivation, workload, and kebele level production numbers for the previous year. After the baseline, we conducted extensive training on the RCT design for woreda-level Bureau of Agriculture functionaries and DAs in collaboration with Digital Green.
After these trainings, the intervention was implemented by woreda-level subject matter specialists and DAs with support from Digital Green throughout the main production period of the meher season (June-September 2017). We also conducted a rapid assessment of the implementation process during the initial implementation in order to provide feedback to Digital Green on implementation progress. The household survey and DA follow-up survey were conducted January-March 2018, after harvest. The Year 2 household survey and DA follow-up survey follow the same schedule, and are being conducted in January-March 2019, after harvest.